diff --git "a/data.json" "b/data.json" --- "a/data.json" +++ "b/data.json" @@ -1,3248 +1,3314 @@ [ { - "metrics.accuracy": "ROUGEL: 21.658590925735947 exact_match: 90.0034579748511 TOKENS_PER_SAMPLE: 648.3", - "metrics.result": 428.895, - "metrics.result_per_accelerator": 53.611875, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", - "software.framework.ai": null, - "software.framework.cuda": "12.8", - "software.framework.cudnn": null, - "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", - "submission.availability": "available", - "submission.division": "open", - "submission.organization": "HPE", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 192, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", - "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", - "system.number_of_nodes": 1, - "system.type": "datacenter" - }, - { - "metrics.accuracy": "ROUGE1: 30.6202 ROUGE2: 13.9221 ROUGEL: 18.9101 TOKENS_PER_SAMPLE: 581.8", - "metrics.result": 2631.93, - "metrics.result_per_accelerator": 2631.93, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5321 ROUGE2: 22.1191 ROUGEL: 28.7225 TOKENS_PER_SAMPLE: 293.7", + "metrics.result": 3062.72, + "metrics.result_per_accelerator": 3062.72, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", - "model.name": "DeepSeek-R1-Distill-Llama-8B", - "model.number_of_parameters": 8.0, - "model.weight_data_types": "bfloat16", - "software.framework": "vLLM v0.7.3", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.2", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS (5.15.0-131-generic)", + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "FlexAI", - "submission.scenario": "Server", + "submission.division": "closed", + "submission.organization": "AMD", + "submission.scenario": "Offline", "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": null, + "system.accelerator.memory_capacity": 192.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100 80GB HBM3", + "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", "system.accelerator.total_count": 1, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d cache: 6.3 MiB (200 instances), L1i cache: 6.3 MiB (200 instances), L2 cache: 800 MiB (200 instances), L3 cache: 3.1 GiB (200 instances)", - "system.cpu.core_count": 52, + "system.accelerator.vendor": "AMD", + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel Xeon Processor (SapphireRapids)", + "system.cpu.model": "2xAMD EPYC 9374F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": null, - "system.memory.configuration": "undefined", - "system.name": "flexbench test node 0ef307db09d34a91 with 8xH100", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", + "system.name": "Supermicro AS-8125GS-TNMR2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.246 ROUGE2: 22.4512 ROUGEL: 27.9398 TOKENS_PER_SAMPLE: 335.5", - "metrics.result": 134.714, - "metrics.result_per_accelerator": 134.714, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5105 ROUGE2: 22.0812 ROUGEL: 28.727 TOKENS_PER_SAMPLE: 292.7", + "metrics.result": 2520.27, + "metrics.result_per_accelerator": 2520.27, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", - "model.name": "Llama-3.3-70B-Instruct-FP8-dynamic", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "bfloat16", - "software.framework": "vLLM v0.7.3", + "model.weight_data_types": "fp8", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.2", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS (5.15.0-131-generic)", + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "FlexAI", + "submission.division": "closed", + "submission.organization": "AMD", "submission.scenario": "Server", "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": null, + "system.accelerator.memory_capacity": 192.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100 80GB HBM3", + "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", "system.accelerator.total_count": 1, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d cache: 6.3 MiB (200 instances), L1i cache: 6.3 MiB (200 instances), L2 cache: 800 MiB (200 instances), L3 cache: 3.1 GiB (200 instances)", - "system.cpu.core_count": 52, + "system.accelerator.vendor": "AMD", + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel Xeon Processor (SapphireRapids)", + "system.cpu.model": "2xAMD EPYC 9374F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": null, - "system.memory.configuration": "undefined", - "system.name": "flexbench test node 0ef307db09d34a91 with 8xH100", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", + "system.name": "Supermicro AS-8125GS-TNMR2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4619 ROUGE2: 22.0574 ROUGEL: 28.6452 TOKENS_PER_SAMPLE: 295.9", - "metrics.result": 1482.31, - "metrics.result_per_accelerator": 370.5775, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5321 ROUGE2: 22.1191 ROUGEL: 28.7225 TOKENS_PER_SAMPLE: 293.7", + "metrics.result": 3062.72, + "metrics.result_per_accelerator": 3062.72, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "Llama-2-70b-chat-hf-FP8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM v0.7.3", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.2", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS (5.15.0-131-generic)", + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "FlexAI", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": null, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100 80GB HBM3", - "system.accelerator.total_count": 4, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d cache: 6.3 MiB (200 instances), L1i cache: 6.3 MiB (200 instances), L2 cache: 800 MiB (200 instances), L3 cache: 3.1 GiB (200 instances)", - "system.cpu.core_count": 52, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel Xeon Processor (SapphireRapids)", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": null, - "system.memory.configuration": "undefined", - "system.name": "flexbench test node 0ef307db09d34a91 with 8xH100", - "system.number_of_nodes": 1, - "system.type": "datacenter" - }, - { - "metrics.accuracy": "ROUGE1: 44.8294 ROUGE2: 22.3814 ROUGEL: 29.1878 TOKENS_PER_SAMPLE: 271.8", - "metrics.result": 573.765, - "metrics.result_per_accelerator": 573.765, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.8, CUDA 12.8", - "software.framework.ai": null, - "software.framework.cuda": "12.8", - "software.framework.cudnn": null, - "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Jetson L4T", - "submission.availability": "preview", - "submission.division": "open", - "submission.organization": "NVIDIA", + "submission.division": "closed", + "submission.organization": "AMD", "submission.scenario": "Offline", "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": null, - "system.accelerator.memory_config": "LPDDR5X", - "system.accelerator.name": "NVIDIA Jetson AGX Thor 128G", + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", "system.accelerator.total_count": 1, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 14, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "14-core ARM Poseidon-AE CPU", + "system.cpu.model": "2xAMD EPYC 9374F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "ARM", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 128.0, - "system.memory.configuration": "128GB 256-bit LPDDR5X", - "system.name": "NVIDIA Jetson AGX Thor Developer Kit 128G (TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", + "system.name": "Supermicro AS-8125GS-TNMR2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.8294 ROUGE2: 22.3814 ROUGEL: 29.1878 TOKENS_PER_SAMPLE: 271.8", - "metrics.result": 573.765, - "metrics.result_per_accelerator": 573.765, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5105 ROUGE2: 22.0812 ROUGEL: 28.727 TOKENS_PER_SAMPLE: 292.7", + "metrics.result": 2520.27, + "metrics.result_per_accelerator": 2520.27, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp8", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.2", "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Jetson L4T", - "submission.availability": "preview", - "submission.division": "open", - "submission.organization": "NVIDIA", - "submission.scenario": "Offline", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "AMD", + "submission.scenario": "Server", "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": null, - "system.accelerator.memory_config": "LPDDR5X", - "system.accelerator.name": "NVIDIA Jetson AGX Thor 128G", + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", "system.accelerator.total_count": 1, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 14, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "14-core ARM Poseidon-AE CPU", + "system.cpu.model": "2xAMD EPYC 9374F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "ARM", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 128.0, - "system.memory.configuration": "128GB 256-bit LPDDR5X", - "system.name": "NVIDIA Jetson AGX Thor Developer Kit 128G (TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", + "system.name": "Supermicro AS-8125GS-TNMR2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 47.6186 ROUGE2: 25.2204 ROUGEL: 31.4595 TOKENS_PER_SAMPLE: 296.9", - "metrics.result": 16162.8, - "metrics.result_per_accelerator": 2020.35, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5633 ROUGE2: 22.1252 ROUGEL: 28.7479 TOKENS_PER_SAMPLE: 292.7", + "metrics.result": 23514.8, + "metrics.result_per_accelerator": 2939.35, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama3_1-70b-fp8_dyn", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "SGLang v0.4.3", + "model.weight_data_types": "fp8", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.2", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": null, + "software.framework.vllm": "0.4.3", "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "AMD", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 192.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", "system.accelerator.total_count": 8, "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 40, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.model": "2xAMD EPYC 9374F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, SGLang v0.4.3)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", + "system.name": "Supermicro AS-8125GS-TNMR2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 18.9480793556203 exact_match: 90.14579086697552 TOKENS_PER_SAMPLE: 641.1", - "metrics.result": 243.449, - "metrics.result_per_accelerator": 30.431125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5522 ROUGE2: 22.1184 ROUGEL: 28.7592 TOKENS_PER_SAMPLE: 293.5", + "metrics.result": 21028.2, + "metrics.result_per_accelerator": 2628.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "deepseek-v3", - "model.number_of_parameters": 671.0, - "model.weight_data_types": "UNSET", - "software.framework": "SGLang v0.4.3", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.2", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": null, + "software.framework.vllm": "0.4.3", "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", - "submission.scenario": "Offline", + "submission.division": "closed", + "submission.organization": "AMD", + "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 192.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", "system.accelerator.total_count": 8, "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 40, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.model": "2xAMD EPYC 9374F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, SGLang v0.4.3)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", + "system.name": "Supermicro AS-8125GS-TNMR2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5375 ROUGE2: 22.202 ROUGEL: 28.7796 TOKENS_PER_SAMPLE: 299.2", - "metrics.result": 16229.0, - "metrics.result_per_accelerator": 2028.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5633 ROUGE2: 22.1252 ROUGEL: 28.7479 TOKENS_PER_SAMPLE: 292.7", + "metrics.result": 23514.8, + "metrics.result_per_accelerator": 2939.35, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-fp8_dyn", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "UNSET", - "software.framework": "vLLM v0.7.3", + "model.weight_data_types": "fp8", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.2", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "AMD", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.accelerator.vendor": "AMD", + "system.cpu.caches": null, "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "2xAMD EPYC 9374F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, vLLM v0.7.3)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", + "system.name": "Supermicro AS-8125GS-TNMR2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6113 ROUGE2: 22.2456 ROUGEL: 28.8393 TOKENS_PER_SAMPLE: 298.1", - "metrics.result": 14426.4, - "metrics.result_per_accelerator": 1803.3, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5522 ROUGE2: 22.1184 ROUGEL: 28.7592 TOKENS_PER_SAMPLE: 293.5", + "metrics.result": 21028.2, + "metrics.result_per_accelerator": 2628.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-fp8_dyn", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "UNSET", - "software.framework": "vLLM v0.7.3", + "model.weight_data_types": "fp8", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.2", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "AMD", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.accelerator.vendor": "AMD", + "system.cpu.caches": null, "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "2xAMD EPYC 9374F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, vLLM v0.7.3)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", + "system.name": "Supermicro AS-8125GS-TNMR2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.59129621595158 exact_match: 90.09433818663138 TOKENS_PER_SAMPLE: 641.0", - "metrics.result": 188.604, - "metrics.result_per_accelerator": 23.5755, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5633 ROUGE2: 22.1252 ROUGEL: 28.7479 TOKENS_PER_SAMPLE: 292.7", + "metrics.result": 24109.8, + "metrics.result_per_accelerator": 3013.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp16", - "software.framework": "vLLM v0.7.3", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.2", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", - "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.6 LTS (Jammy Jellyfish)", + "submission.availability": "preview", + "submission.division": "closed", + "submission.organization": "AMD", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 192.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", "system.accelerator.total_count": 8, "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 40, + "system.cpu.core_count": null, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.model": "2xAMD EPYC TURIN", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, vLLM v0.7.3)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB Samsung M321R8GA0PB1-CCPXC", + "system.name": "Supermicro AS-8125GS-TNMR2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.66444823059425 exact_match: 90.10726671078757 TOKENS_PER_SAMPLE: 632.1", - "metrics.result": 278.649, - "metrics.result_per_accelerator": 34.831125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5555 ROUGE2: 22.0993 ROUGEL: 28.7204 TOKENS_PER_SAMPLE: 292.8", + "metrics.result": 22020.9, + "metrics.result_per_accelerator": 2752.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b-fp8_dyn", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp16", - "software.framework": "vLLM v0.7.3", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.2", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", - "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", - "submission.scenario": "Offline", + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.6 LTS (Jammy Jellyfish)", + "submission.availability": "preview", + "submission.division": "closed", + "submission.organization": "AMD", + "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 192.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", "system.accelerator.total_count": 8, "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 40, + "system.cpu.core_count": null, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.model": "2xAMD EPYC TURIN", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, vLLM v0.7.3)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB Samsung M321R8GA0PB1-CCPXC", + "system.name": "Supermicro AS-8125GS-TNMR2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4974 ROUGE2: 22.1155 ROUGEL: 28.7195 TOKENS_PER_SAMPLE: 298.1", - "metrics.result": 17128.9, - "metrics.result_per_accelerator": 2141.1125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5633 ROUGE2: 22.1252 ROUGEL: 28.7479 TOKENS_PER_SAMPLE: 292.7", + "metrics.result": 24109.8, + "metrics.result_per_accelerator": 3013.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-fp8_dyn", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "vLLM v0.7.3", + "model.weight_data_types": "fp8", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.2", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", - "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.6 LTS (Jammy Jellyfish)", + "submission.availability": "preview", + "submission.division": "closed", + "submission.organization": "AMD", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 192.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", "system.accelerator.total_count": 8, "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 40, + "system.cpu.core_count": null, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.model": "2xAMD EPYC TURIN", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, vLLM v0.7.3)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB Samsung M321R8GA0PB1-CCPXC", + "system.name": "Supermicro AS-8125GS-TNMR2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5066 ROUGE2: 22.1409 ROUGEL: 28.7316 TOKENS_PER_SAMPLE: 297.1", - "metrics.result": 12989.0, - "metrics.result_per_accelerator": 1623.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5555 ROUGE2: 22.0993 ROUGEL: 28.7204 TOKENS_PER_SAMPLE: 292.8", + "metrics.result": 22020.9, + "metrics.result_per_accelerator": 2752.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-fp8_dyn", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "vLLM v0.7.3", + "model.weight_data_types": "fp8", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.2", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", - "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.6 LTS (Jammy Jellyfish)", + "submission.availability": "preview", + "submission.division": "closed", + "submission.organization": "AMD", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 192.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", "system.accelerator.total_count": 8, "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 40, + "system.cpu.core_count": null, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.model": "2xAMD EPYC TURIN", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, vLLM v0.7.3)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB Samsung M321R8GA0PB1-CCPXC", + "system.name": "Supermicro AS-8125GS-TNMR2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5183 ROUGE2: 22.1357 ROUGEL: 28.7267 TOKENS_PER_SAMPLE: 297.0", - "metrics.result": 21372.3, - "metrics.result_per_accelerator": 2671.5375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 6985.51, + "metrics.result_per_accelerator": 1746.3775, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-fp8_dyn", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp16", - "software.framework": "vLLM v0.6.4 FlashInfer", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.caches": null, + "system.cpu.core_count": 96, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9654 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, vLLM v0.6.4)", + "system.memory.capacity": 768.0, + "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", + "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5421 ROUGE2: 22.1749 ROUGEL: 28.8061 TOKENS_PER_SAMPLE: 295.1", - "metrics.result": 19618.1, - "metrics.result_per_accelerator": 2452.2625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1008 ROUGE2: 20.1119 ROUGEL: 29.9747 GEN_LEN: 4114209", + "metrics.result": 6070.26, + "metrics.result_per_accelerator": 1517.565, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-fp8_dyn", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp16", - "software.framework": "vLLM v0.6.4 FlashInfer", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.caches": null, + "system.cpu.core_count": 96, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9654 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, vLLM v0.6.4)", + "system.memory.capacity": 768.0, + "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", + "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5823 ROUGE2: 22.2287 ROUGEL: 28.8178 TOKENS_PER_SAMPLE: 296.5", - "metrics.result": 15906.1, - "metrics.result_per_accelerator": 1988.2625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 6985.51, + "metrics.result_per_accelerator": 1746.3775, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-fp8_dyn", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp16", - "software.framework": "vLLM v0.6.4", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.caches": null, + "system.cpu.core_count": 96, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9654 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, vLLM v0.6.4)", + "system.memory.capacity": 768.0, + "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", + "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5362 ROUGE2: 22.176 ROUGEL: 28.773 TOKENS_PER_SAMPLE: 295.7", - "metrics.result": 13716.7, - "metrics.result_per_accelerator": 1714.5875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1008 ROUGE2: 20.1119 ROUGEL: 29.9747 GEN_LEN: 4114209", + "metrics.result": 6070.26, + "metrics.result_per_accelerator": 1517.565, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-fp8_dyn", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp16", - "software.framework": "vLLM v0.6.4", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.caches": null, + "system.cpu.core_count": 96, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9654 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, vLLM v0.6.4)", + "system.memory.capacity": 768.0, + "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", + "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4973 ROUGE2: 22.1394 ROUGEL: 28.7255 TOKENS_PER_SAMPLE: 301.3", - "metrics.result": 18561.4, - "metrics.result_per_accelerator": 2320.175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5664 ROUGE2: 22.1601 ROUGEL: 28.7722 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 6673.95, + "metrics.result_per_accelerator": 1668.4875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-fp8_dyn", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp16", - "software.framework": "vLLM v0.6.4", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 40, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 96, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.model": "AMD EPYC 9654 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, vLLM v0.6.4)", + "system.memory.capacity": 768.0, + "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", + "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5398 ROUGE2: 22.1484 ROUGEL: 28.725 TOKENS_PER_SAMPLE: 299.5", - "metrics.result": 13916.2, - "metrics.result_per_accelerator": 1739.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5664 ROUGE2: 22.16 ROUGEL: 28.7714 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 5788.29, + "metrics.result_per_accelerator": 1447.0725, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-fp8_dyn", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp16", - "software.framework": "vLLM v0.6.4", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 40, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 96, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.model": "AMD EPYC 9654 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, vLLM v0.6.4)", + "system.memory.capacity": 768.0, + "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", + "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 47.5723 ROUGE2: 25.2003 ROUGEL: 31.4338 TOKENS_PER_SAMPLE: 299.6", - "metrics.result": 24624.1, - "metrics.result_per_accelerator": 3078.0125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5664 ROUGE2: 22.1601 ROUGEL: 28.7722 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 6673.95, + "metrics.result_per_accelerator": 1668.4875, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama3_1-70b-fp8_dyn", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp16", - "software.framework": "SGLang v0.4.3", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.caches": null, + "system.cpu.core_count": 96, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9654 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, SGLang v0.4.3)", + "system.memory.capacity": 768.0, + "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", + "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 47.589 ROUGE2: 25.2681 ROUGEL: 31.545 TOKENS_PER_SAMPLE: 292.1", - "metrics.result": 27687.9, - "metrics.result_per_accelerator": 3460.9875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5664 ROUGE2: 22.16 ROUGEL: 28.7714 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 5788.29, + "metrics.result_per_accelerator": 1447.0725, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama3_1-70b-fp8_pre", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "SGLang v0.4.3", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "submission.division": "closed", + "submission.organization": "ASUSTeK", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.caches": null, + "system.cpu.core_count": 96, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9654 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, SGLang v0.4.3)", + "system.memory.capacity": 768.0, + "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", + "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 47.6531 ROUGE2: 25.2981 ROUGEL: 31.5596 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 26074.4, - "metrics.result_per_accelerator": 3259.3, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1261 ROUGE2: 20.1655 ROUGEL: 30.025 GEN_LEN: 4101272", + "metrics.result": 13078.5, + "metrics.result_per_accelerator": 1634.8125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama3_1-70b-fp8_pre", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "SGLang v0.4.3", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", - "submission.scenario": "Server", + "submission.division": "closed", + "submission.organization": "ASUSTeK", + "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, SGLang v0.4.3)", + "system.memory.configuration": "16x 64GB ", + "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 47.5684 ROUGE2: 25.2712 ROUGEL: 31.533 TOKENS_PER_SAMPLE: 292.8", - "metrics.result": 27949.8, - "metrics.result_per_accelerator": 3493.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.126 ROUGE2: 20.1652 ROUGEL: 30.026 GEN_LEN: 4101320", + "metrics.result": 7986.46, + "metrics.result_per_accelerator": 998.3075, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama3_1-70b-interactive-fp8_pre", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "SGLang v0.4.3", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", - "submission.scenario": "Offline", + "submission.division": "closed", + "submission.organization": "ASUSTeK", + "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, SGLang v0.4.3)", + "system.memory.configuration": "16x 64GB ", + "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 47.6193 ROUGE2: 25.2818 ROUGEL: 31.5351 TOKENS_PER_SAMPLE: 291.8", - "metrics.result": 12192.4, - "metrics.result_per_accelerator": 1524.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1261 ROUGE2: 20.1655 ROUGEL: 30.025 GEN_LEN: 4101272", + "metrics.result": 13078.5, + "metrics.result_per_accelerator": 1634.8125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama3_1-70b-interactive-fp8_pre", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "SGLang v0.4.3", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", - "submission.scenario": "Server", + "submission.division": "closed", + "submission.organization": "ASUSTeK", + "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, SGLang v0.4.3)", + "system.memory.configuration": "16x 64GB ", + "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.634266581485935 exact_match: 90.12628722700197 TOKENS_PER_SAMPLE: 656.3", - "metrics.result": 326.537, - "metrics.result_per_accelerator": 40.817125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.126 ROUGE2: 20.1652 ROUGEL: 30.026 GEN_LEN: 4101320", + "metrics.result": 7986.46, + "metrics.result_per_accelerator": 998.3075, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b-fp8_dyn", - "model.number_of_parameters": 405.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp16", - "software.framework": "SGLang v0.4.3", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", - "submission.scenario": "Offline", + "submission.division": "closed", + "submission.organization": "ASUSTeK", + "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, SGLang v0.4.3)", + "system.memory.configuration": "16x 64GB ", + "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 46.5248 ROUGE2: 24.2311 ROUGEL: 30.2319 TOKENS_PER_SAMPLE: 323.9", - "metrics.result": 30530.1, - "metrics.result_per_accelerator": 3816.2625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.566 ROUGE2: 22.16 ROUGEL: 28.7719 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 9281.5, + "metrics.result_per_accelerator": 1160.1875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama3_1-70b-interactive-fp8_nim", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp16", - "software.framework": "NIM v1.5.0", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, NIM v1.5.0)", + "system.memory.configuration": "16x 64GB ", + "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 46.5348 ROUGE2: 24.2605 ROUGEL: 30.246 TOKENS_PER_SAMPLE: 324.7", - "metrics.result": 15960.1, - "metrics.result_per_accelerator": 1995.0125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5654 ROUGE2: 22.1595 ROUGEL: 28.7714 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 8094.27, + "metrics.result_per_accelerator": 1011.78375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama3_1-70b-interactive-fp8_nim", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp16", - "software.framework": "NIM v1.5.0", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, NIM v1.5.0)", + "system.memory.configuration": "16x 64GB ", + "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 46.5273 ROUGE2: 24.2307 ROUGEL: 30.2329 TOKENS_PER_SAMPLE: 323.8", - "metrics.result": 30042.8, - "metrics.result_per_accelerator": 3755.35, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.566 ROUGE2: 22.16 ROUGEL: 28.7719 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 9281.5, + "metrics.result_per_accelerator": 1160.1875, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama3_1-70b-fp8_nim", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp16", - "software.framework": "NIM v1.5.0", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, NIM v1.5.0)", + "system.memory.configuration": "16x 64GB ", + "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 46.5201 ROUGE2: 24.2273 ROUGEL: 30.2319 TOKENS_PER_SAMPLE: 323.9", - "metrics.result": 28421.0, - "metrics.result_per_accelerator": 3552.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5654 ROUGE2: 22.1595 ROUGEL: 28.7714 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 8094.27, + "metrics.result_per_accelerator": 1011.78375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama3_1-70b-fp8_nim", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp16", - "software.framework": "NIM v1.5.0", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Krai", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, NIM v1.5.0)", + "system.memory.configuration": "16x 64GB ", + "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", - "metrics.result": 97403.4, - "metrics.result_per_accelerator": 12175.425, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 19877.6, + "metrics.result_per_accelerator": 2484.7, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Lambda", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", + "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.6315 ROUGE2: 23.4063 ROUGEL: 30.5125 TOKENS_PER_SAMPLE: 144.1 gsm8k_accuracy: 73.06 mbxp_accuracy: 60.38", - "metrics.result": 121886.0, - "metrics.result_per_accelerator": 15235.75, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0994 ROUGE2: 20.1115 ROUGEL: 29.9734 GEN_LEN: 4114014", + "metrics.result": 19226.9, + "metrics.result_per_accelerator": 2403.3625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Lambda", - "submission.scenario": "Offline", + "submission.division": "closed", + "submission.organization": "ASUSTeK", + "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", + "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", - "metrics.result": 97403.4, - "metrics.result_per_accelerator": 12175.425, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 19877.6, + "metrics.result_per_accelerator": 2484.7, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Lambda", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", + "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.738071137617286 exact_match: 90.02873593646592 TOKENS_PER_SAMPLE: 660.4", - "metrics.result": 564.696, - "metrics.result_per_accelerator": 70.587, - "metrics.units": "Tokens/s", + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0994 ROUGE2: 20.1115 ROUGEL: 29.9734 GEN_LEN: 4114014", + "metrics.result": 19226.9, + "metrics.result_per_accelerator": 2403.3625, + "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Supermicro", - "submission.scenario": "Offline", + "submission.division": "closed", + "submission.organization": "ASUSTeK", + "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", + "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 2791.38, - "metrics.result_per_accelerator": 2791.38, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5663 ROUGE2: 22.1602 ROUGEL: 28.7722 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 24323.6, + "metrics.result_per_accelerator": 3040.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Supermicro", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 64, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "LPDDR5x", - "system.name": "ARS-111GL-NHR (1x GH200-96GB_aarch64)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", + "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 2791.38, - "metrics.result_per_accelerator": 2791.38, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5663 ROUGE2: 22.1597 ROUGEL: 28.7718 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 20605.3, + "metrics.result_per_accelerator": 2575.6625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Supermicro", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, + "submission.division": "closed", + "submission.organization": "ASUSTeK", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 64, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "LPDDR5x", - "system.name": "ARS-111GL-NHR (1x GH200-96GB_aarch64)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", + "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5498 ROUGE2: 22.134 ROUGEL: 28.7842 TOKENS_PER_SAMPLE: 291.4", - "metrics.result": 8231.9, - "metrics.result_per_accelerator": 4115.95, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5663 ROUGE2: 22.1602 ROUGEL: 28.7722 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 24323.6, + "metrics.result_per_accelerator": 3040.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Red Hat 9.4 + OpenShift 4.15", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Supermicro", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "ARS-221GL-NHIR (2x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", + "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5491 ROUGE2: 22.1346 ROUGEL: 28.7871 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 7538.8, - "metrics.result_per_accelerator": 3769.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5663 ROUGE2: 22.1597 ROUGEL: 28.7718 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 20605.3, + "metrics.result_per_accelerator": 2575.6625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Red Hat 9.4 + OpenShift 4.15", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Supermicro", + "submission.division": "closed", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "ARS-221GL-NHIR (2x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", + "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5498 ROUGE2: 22.134 ROUGEL: 28.7842 TOKENS_PER_SAMPLE: 291.4", - "metrics.result": 8231.9, - "metrics.result_per_accelerator": 4115.95, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0843 ROUGE2: 20.0853 ROUGEL: 29.9346 GEN_LEN: 4114897", + "metrics.result": 1747.67, + "metrics.result_per_accelerator": 873.835, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Red Hat 9.4 + OpenShift 4.15", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Supermicro", + "submission.division": "closed", + "submission.organization": "Cisco", "submission.scenario": "Offline", "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel Xeon Gold 6448H", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "ARS-221GL-NHIR (2x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", + "system.name": "Cisco UCS C240 M7 (2x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5491 ROUGE2: 22.1346 ROUGEL: 28.7871 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 7538.8, - "metrics.result_per_accelerator": 3769.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.0866 ROUGEL: 29.9354 GEN_LEN: 4114953", + "metrics.result": 1734.03, + "metrics.result_per_accelerator": 867.015, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Red Hat 9.4 + OpenShift 4.15", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "Supermicro", + "submission.division": "closed", + "submission.organization": "Cisco", "submission.scenario": "Server", "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel Xeon Gold 6448H", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "ARS-221GL-NHIR (2x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", + "system.name": "Cisco UCS C240 M7 (2x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4322 ROUGE2: 22.0403 ROUGEL: 28.6357 TOKENS_PER_SAMPLE: 294.3", - "metrics.result": 31238.0, - "metrics.result_per_accelerator": 3904.75, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0362 ROUGE2: 20.0871 ROUGEL: 29.9403 GEN_LEN: 4123417", + "metrics.result": 3268.94, + "metrics.result_per_accelerator": 1634.47, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Cisco", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 2, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9684X 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", + "system.name": "Cisco UCS C245 M8 (2x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4297 ROUGE2: 22.0409 ROUGEL: 28.6367 TOKENS_PER_SAMPLE: 295.4", - "metrics.result": 30457.7, - "metrics.result_per_accelerator": 3807.2125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0354 ROUGE2: 20.0871 ROUGEL: 29.9398 GEN_LEN: 4123289", + "metrics.result": 3258.83, + "metrics.result_per_accelerator": 1629.415, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Cisco", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 2, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9684X 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", + "system.name": "Cisco UCS C245 M8 (2x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20796.0, - "metrics.result_per_accelerator": 2599.5, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0605 ROUGE2: 20.1151 ROUGEL: 29.9703 GEN_LEN: 4112035", + "metrics.result": 1729.6, + "metrics.result_per_accelerator": 864.8, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Cisco", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9684X 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", + "system.name": "Cisco UCS C245 M8 (2x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 19363.3, - "metrics.result_per_accelerator": 2420.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0621 ROUGE2: 20.1162 ROUGEL: 29.9713 GEN_LEN: 4112159", + "metrics.result": 1725.54, + "metrics.result_per_accelerator": 862.77, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Cisco", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9684X 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", + "system.name": "Cisco UCS C245 M8 (2x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20993.9, - "metrics.result_per_accelerator": 2624.2375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0605 ROUGE2: 20.1151 ROUGEL: 29.9703 GEN_LEN: 4112035", + "metrics.result": 1746.46, + "metrics.result_per_accelerator": 873.23, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Cisco", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", + "system.name": "Cisco UCS X210c M7 (2x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 19363.2, - "metrics.result_per_accelerator": 2420.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0605 ROUGE2: 20.1151 ROUGEL: 29.9703 GEN_LEN: 4112037", + "metrics.result": 1725.98, + "metrics.result_per_accelerator": 862.99, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Cisco", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", + "system.name": "Cisco UCS X210c M7 (2x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4546 ROUGE2: 22.0578 ROUGEL: 28.6551 TOKENS_PER_SAMPLE: 294.3", - "metrics.result": 23389.3, - "metrics.result_per_accelerator": 2923.6625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0605 ROUGE2: 20.1151 ROUGEL: 29.9703 GEN_LEN: 4112035", + "metrics.result": 1746.46, + "metrics.result_per_accelerator": 873.23, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Cisco", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", + "system.name": "Cisco UCS X210c M7 (2x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5336 ROUGE2: 22.0916 ROUGEL: 28.7043 TOKENS_PER_SAMPLE: 294.3", - "metrics.result": 14008.5, - "metrics.result_per_accelerator": 1751.0625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0605 ROUGE2: 20.1151 ROUGEL: 29.9703 GEN_LEN: 4112037", + "metrics.result": 1725.98, + "metrics.result_per_accelerator": 862.99, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Cisco", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", + "system.name": "Cisco UCS X210c M7 (2x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4546 ROUGE2: 22.0578 ROUGEL: 28.6551 TOKENS_PER_SAMPLE: 294.3", - "metrics.result": 23385.0, - "metrics.result_per_accelerator": 2923.125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 42.9363 ROUGE2: 20.1275 ROUGEL: 29.9464 GEN_LEN: 4048420", + "metrics.result": 64.0078, + "metrics.result_per_accelerator": 64.0078, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "Jetpack 6.0, TensorRT 10.1, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, - "software.framework.jetpack": null, + "software.framework.jetpack": "6.0", "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.1", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Jetson r36.3.0 L4T", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "ConnectTechInc", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "LPDDR5", + "system.accelerator.name": "NVIDIA Jetson AGX Orin 64G", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.caches": null, + "system.cpu.core_count": 12, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "12-core ARM Cortex-A78AE CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "ARM", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 64.0, + "system.memory.configuration": "64GB 256-bit LPDDR5", + "system.name": "NVIDIA Jetson AGX Orin 64G (TensorRT) + CTI Forge Carrier (AGX201)", "system.number_of_nodes": 1, - "system.type": "datacenter" + "system.type": "edge" }, { - "metrics.accuracy": "ROUGE1: 44.5144 ROUGE2: 22.0732 ROUGEL: 28.686 TOKENS_PER_SAMPLE: 294.4", - "metrics.result": 13987.6, - "metrics.result_per_accelerator": 1748.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0343 ROUGE2: 20.0912 ROUGEL: 29.9254 GEN_LEN: 4102194", + "metrics.result": 7191.41, + "metrics.result_per_accelerator": 1797.8525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "submission.organization": "Dell", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator": "PCIe", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB 4400 MT/s", + "system.name": "Dell PowerEdge R760xa (4xH100 NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4347 ROUGE2: 22.0435 ROUGEL: 28.6357 TOKENS_PER_SAMPLE: 294.3", - "metrics.result": 31252.1, - "metrics.result_per_accelerator": 3906.5125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0347 ROUGE2: 20.0913 ROUGEL: 29.9262 GEN_LEN: 4102176", + "metrics.result": 6337.64, + "metrics.result_per_accelerator": 1584.41, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "submission.organization": "Dell", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator": "PCIe", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB 4400 MT/s", + "system.name": "Dell PowerEdge R760xa (4xH100 NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4184 ROUGE2: 22.033 ROUGEL: 28.6265 TOKENS_PER_SAMPLE: 295.6", - "metrics.result": 30500.8, - "metrics.result_per_accelerator": 3812.6, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1259 ROUGE2: 20.1653 ROUGEL: 30.0252 GEN_LEN: 4101368", + "metrics.result": 3317.08, + "metrics.result_per_accelerator": 1658.54, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, + "submission.organization": "Dell", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 2, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator": "PCIe", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB 4400 MT/s", + "system.name": "Dell PowerEdge R760 (2xH100_PCIe_80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5458 ROUGE2: 22.1332 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 30297.5, - "metrics.result_per_accelerator": 3787.1875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1239 ROUGE2: 20.1642 ROUGEL: 30.0234 GEN_LEN: 4101242", + "metrics.result": 2759.33, + "metrics.result_per_accelerator": 1379.665, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "submission.organization": "Dell", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", - "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", + "system.interconnect.accelerator": "PCIe", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB 4400 MT/s", + "system.name": "Dell PowerEdge R760 (2xH100_PCIe_80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5621 ROUGE2: 22.1437 ROUGEL: 28.7928 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 27638.9, - "metrics.result_per_accelerator": 3454.8625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1259 ROUGE2: 20.1653 ROUGEL: 30.0252 GEN_LEN: 4101368", + "metrics.result": 3317.08, + "metrics.result_per_accelerator": 1658.54, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 192, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", - "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", - "system.number_of_nodes": 1, - "system.type": "datacenter" - }, - { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 17882.9, - "metrics.result_per_accelerator": 2235.3625, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", - "software.framework.ai": null, - "software.framework.cuda": "12.8", - "software.framework.cudnn": null, - "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", - "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", - "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", + "system.interconnect.accelerator": "PCIe", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB 4400 MT/s", + "system.name": "Dell PowerEdge R760 (2xH100_PCIe_80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 18065.1, - "metrics.result_per_accelerator": 2258.1375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1239 ROUGE2: 20.1642 ROUGEL: 30.0234 GEN_LEN: 4101242", + "metrics.result": 2759.33, + "metrics.result_per_accelerator": 1379.665, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", - "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", + "system.interconnect.accelerator": "PCIe", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB 4400 MT/s", + "system.name": "Dell PowerEdge R760 (2xH100_PCIe_80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 17882.9, - "metrics.result_per_accelerator": 2235.3625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 9966.79, + "metrics.result_per_accelerator": 2491.6975, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", - "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 18065.1, - "metrics.result_per_accelerator": 2258.1375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0994 ROUGE2: 20.1104 ROUGEL: 29.9725 GEN_LEN: 4114148", + "metrics.result": 9990.24, + "metrics.result_per_accelerator": 2497.56, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", - "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3505 ROUGE2: 23.2113 ROUGEL: 30.312 TOKENS_PER_SAMPLE: 144.9 gsm8k_accuracy: 73.02 mbxp_accuracy: 59.82", - "metrics.result": 52205.9, - "metrics.result_per_accelerator": 6525.7375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 9966.79, + "metrics.result_per_accelerator": 2491.6975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", - "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4369 ROUGE2: 23.2825 ROUGEL: 30.4229 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.48 mbxp_accuracy: 59.94", - "metrics.result": 50123.9, - "metrics.result_per_accelerator": 6265.4875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0994 ROUGE2: 20.1104 ROUGEL: 29.9725 GEN_LEN: 4114148", + "metrics.result": 9990.24, + "metrics.result_per_accelerator": 2497.56, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", - "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5458 ROUGE2: 22.1332 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 30297.5, - "metrics.result_per_accelerator": 3787.1875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5354 ROUGE2: 22.0994 ROUGEL: 28.7324 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 10699.6, + "metrics.result_per_accelerator": 2674.9, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", - "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5621 ROUGE2: 22.1437 ROUGEL: 28.7928 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 27638.9, - "metrics.result_per_accelerator": 3454.8625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5352 ROUGE2: 22.0997 ROUGEL: 28.7319 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 9522.2, + "metrics.result_per_accelerator": 2380.55, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", - "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4331 ROUGE2: 22.0455 ROUGEL: 28.6365 TOKENS_PER_SAMPLE: 294.2", - "metrics.result": 31174.7, - "metrics.result_per_accelerator": 3896.8375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5354 ROUGE2: 22.0994 ROUGEL: 28.7324 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 10699.6, + "metrics.result_per_accelerator": 2674.9, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4303 ROUGE2: 22.039 ROUGEL: 28.6356 TOKENS_PER_SAMPLE: 295.3", - "metrics.result": 30498.9, - "metrics.result_per_accelerator": 3812.3625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5352 ROUGE2: 22.0997 ROUGEL: 28.7319 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 9522.2, + "metrics.result_per_accelerator": 2380.55, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20716.5, - "metrics.result_per_accelerator": 2589.5625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3721 ROUGE2: 23.2543 ROUGEL: 30.3259 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.12 mbxp_accuracy: 59.84", + "metrics.result": 26824.4, + "metrics.result_per_accelerator": 6706.1, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20795.3, - "metrics.result_per_accelerator": 2599.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3594 ROUGE2: 23.2482 ROUGEL: 30.3186 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.08 mbxp_accuracy: 59.9", + "metrics.result": 25707.0, + "metrics.result_per_accelerator": 6426.75, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20925.3, - "metrics.result_per_accelerator": 2615.6625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1787 ROUGEL: 28.7997 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 10594.1, + "metrics.result_per_accelerator": 2648.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE9640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20991.3, - "metrics.result_per_accelerator": 2623.9125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6141 ROUGE2: 22.1791 ROUGEL: 28.7992 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 8937.82, + "metrics.result_per_accelerator": 2234.455, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE9640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5218 ROUGE2: 22.1082 ROUGEL: 28.731 TOKENS_PER_SAMPLE: 293.1", - "metrics.result": 23435.4, - "metrics.result_per_accelerator": 2929.425, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1787 ROUGEL: 28.7997 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 10594.1, + "metrics.result_per_accelerator": 2648.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE9640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5203 ROUGE2: 22.0752 ROUGEL: 28.6917 TOKENS_PER_SAMPLE: 294.2", - "metrics.result": 14004.9, - "metrics.result_per_accelerator": 1750.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6141 ROUGE2: 22.1791 ROUGEL: 28.7992 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 8937.82, + "metrics.result_per_accelerator": 2234.455, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.6", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB DDR5", + "system.name": "Dell PowerEdge XE9640 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5218 ROUGE2: 22.1082 ROUGEL: 28.731 TOKENS_PER_SAMPLE: 293.1", - "metrics.result": 23653.7, - "metrics.result_per_accelerator": 2956.7125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 24086.7, + "metrics.result_per_accelerator": 3010.8375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, @@ -3250,45 +3316,48 @@ "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5263 ROUGE2: 22.0802 ROUGEL: 28.6949 TOKENS_PER_SAMPLE: 294.6", - "metrics.result": 14014.1, - "metrics.result_per_accelerator": 1751.7625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6146 ROUGE2: 22.1791 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 21589.2, + "metrics.result_per_accelerator": 2698.65, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, @@ -3296,45 +3365,48 @@ "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4346 ROUGE2: 22.0435 ROUGEL: 28.6357 TOKENS_PER_SAMPLE: 294.3", - "metrics.result": 31147.4, - "metrics.result_per_accelerator": 3893.425, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 24086.7, + "metrics.result_per_accelerator": 3010.8375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, @@ -3342,45 +3414,48 @@ "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4347 ROUGE2: 22.0423 ROUGEL: 28.6376 TOKENS_PER_SAMPLE: 295.4", - "metrics.result": 30480.6, - "metrics.result_per_accelerator": 3810.075, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6146 ROUGE2: 22.1791 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 21589.2, + "metrics.result_per_accelerator": 2698.65, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, @@ -3388,3081 +3463,3772 @@ "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5024 ROUGE2: 22.0807 ROUGEL: 28.7026 TOKENS_PER_SAMPLE: 293.0", - "metrics.result": 3655.89, - "metrics.result_per_accelerator": 456.98625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 20238.4, + "metrics.result_per_accelerator": 2529.8, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.0, CUDA 12.8", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.0", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", - "system.cpu.core_count": 96, + "system.cpu.caches": null, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.4, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", - "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5208 ROUGE2: 22.0821 ROUGEL: 28.7224 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 3181.65, - "metrics.result_per_accelerator": 397.70625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1012 ROUGE2: 20.113 ROUGEL: 29.9749 GEN_LEN: 4114492", + "metrics.result": 20139.0, + "metrics.result_per_accelerator": 2517.375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.0, CUDA 12.8", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.0", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", - "system.cpu.core_count": 96, + "system.cpu.caches": null, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.4, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", - "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5024 ROUGE2: 22.0807 ROUGEL: 28.7026 TOKENS_PER_SAMPLE: 293.0", - "metrics.result": 3655.89, - "metrics.result_per_accelerator": 456.98625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 20238.4, + "metrics.result_per_accelerator": 2529.8, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.0, CUDA 12.8", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.0", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", - "system.cpu.core_count": 96, + "system.cpu.caches": null, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.4, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", - "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5208 ROUGE2: 22.0821 ROUGEL: 28.7224 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 3181.65, - "metrics.result_per_accelerator": 397.70625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1012 ROUGE2: 20.113 ROUGEL: 29.9749 GEN_LEN: 4114492", + "metrics.result": 20139.0, + "metrics.result_per_accelerator": 2517.375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.0, CUDA 12.8", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.0", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", - "system.cpu.core_count": 96, + "system.cpu.caches": null, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.4, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", - "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5446 ROUGE2: 22.1336 ROUGEL: 28.7828 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 4609.05, - "metrics.result_per_accelerator": 4609.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5883 ROUGE2: 22.1723 ROUGEL: 28.7904 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 32124.3, + "metrics.result_per_accelerator": 4015.5375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.7", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 52, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5531 ROUGE2: 22.1392 ROUGEL: 28.7903 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 4327.68, - "metrics.result_per_accelerator": 4327.68, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5856 ROUGE2: 22.1708 ROUGEL: 28.7886 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 29739.9, + "metrics.result_per_accelerator": 3717.4875, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.7", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 52, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 2845.43, - "metrics.result_per_accelerator": 2845.43, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5883 ROUGE2: 22.1723 ROUGEL: 28.7904 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 32124.3, + "metrics.result_per_accelerator": 4015.5375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.7", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 52, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 2526.03, - "metrics.result_per_accelerator": 2526.03, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5856 ROUGE2: 22.1708 ROUGEL: 28.7886 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 29739.9, + "metrics.result_per_accelerator": 3717.4875, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "software.framework.ai": null, + "software.framework.cuda": "12.4", + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": null, + "software.framework.rocm": null, + "software.framework.sima": null, + "software.framework.tensorrt": "10.2.0", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.3", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Dell", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": null, + "system.cpu.core_count": 52, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5205 ROUGE2: 22.0831 ROUGEL: 28.7052 TOKENS_PER_SAMPLE: 293.6", + "metrics.result": 22677.6, + "metrics.result_per_accelerator": 2834.7, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.0", + "software.framework.ai": null, + "software.framework.cuda": null, + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.0", + "software.framework.sima": null, + "software.framework.tensorrt": null, + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Dell", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", + "system.cpu.caches": null, + "system.cpu.core_count": 40, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x MI300X_192GB, vLLM)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.4783 ROUGE2: 22.0575 ROUGEL: 28.6859 TOKENS_PER_SAMPLE: 293.8", + "metrics.result": 19886.1, + "metrics.result_per_accelerator": 2485.7625, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.0", + "software.framework.ai": null, + "software.framework.cuda": null, + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.0", + "software.framework.sima": null, + "software.framework.tensorrt": null, + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Dell", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", + "system.cpu.caches": null, + "system.cpu.core_count": 40, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x MI300X_192GB, vLLM)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5205 ROUGE2: 22.0831 ROUGEL: 28.7052 TOKENS_PER_SAMPLE: 293.6", + "metrics.result": 22677.6, + "metrics.result_per_accelerator": 2834.7, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.0", + "software.framework.ai": null, + "software.framework.cuda": null, + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.0", + "software.framework.sima": null, + "software.framework.tensorrt": null, + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Dell", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", + "system.cpu.caches": null, + "system.cpu.core_count": 40, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x MI300X_192GB, vLLM)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.4783 ROUGE2: 22.0575 ROUGEL: 28.6859 TOKENS_PER_SAMPLE: 293.8", + "metrics.result": 19886.1, + "metrics.result_per_accelerator": 2485.7625, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.0", + "software.framework.ai": null, + "software.framework.cuda": null, + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": "2.3.0", + "software.framework.rocm": "6.1.0", + "software.framework.sima": null, + "software.framework.tensorrt": null, + "software.framework.vllm": "0.4.3", + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Dell", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", + "system.cpu.caches": null, + "system.cpu.core_count": 40, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x MI300X_192GB, vLLM)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1387 ROUGE2: 20.1794 ROUGEL: 30.0469 GEN_LEN: 4112577", + "metrics.result": 6911.78, + "metrics.result_per_accelerator": 863.9725, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", + "software.framework.ai": null, + "software.framework.cuda": "12.4", + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": null, + "software.framework.rocm": null, + "software.framework.sima": null, + "software.framework.tensorrt": "10.2", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Fujitsu", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 3.4, + "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", + "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1388 ROUGE2: 20.1792 ROUGEL: 30.0469 GEN_LEN: 4112636", + "metrics.result": 6903.45, + "metrics.result_per_accelerator": 862.93125, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", + "software.framework.ai": null, + "software.framework.cuda": "12.4", + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": null, + "software.framework.rocm": null, + "software.framework.sima": null, + "software.framework.tensorrt": "10.2", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Fujitsu", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 3.4, + "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", + "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1387 ROUGE2: 20.1794 ROUGEL: 30.0469 GEN_LEN: 4112577", + "metrics.result": 6911.78, + "metrics.result_per_accelerator": 863.9725, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.7", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Fujitsu", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 3.4, + "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", + "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1388 ROUGE2: 20.1792 ROUGEL: 30.0469 GEN_LEN: 4112636", + "metrics.result": 6903.45, + "metrics.result_per_accelerator": 862.93125, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", + "software.framework.ai": null, + "software.framework.cuda": "12.4", + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": null, + "software.framework.rocm": null, + "software.framework.sima": null, + "software.framework.tensorrt": "10.2", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Fujitsu", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 3.4, + "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", + "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5205 ROUGE2: 22.0796 ROUGEL: 28.6977 TOKENS_PER_SAMPLE: 292.9", + "metrics.result": 3717.74, + "metrics.result_per_accelerator": 464.7175, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", + "software.framework.ai": null, + "software.framework.cuda": "12.4", + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": null, + "software.framework.rocm": null, + "software.framework.sima": null, + "software.framework.tensorrt": "10.2", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Fujitsu", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 3.4, + "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", + "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5132 ROUGE2: 22.0915 ROUGEL: 28.7055 TOKENS_PER_SAMPLE: 293.0", + "metrics.result": 3218.55, + "metrics.result_per_accelerator": 402.31875, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", + "software.framework.ai": null, + "software.framework.cuda": "12.4", + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": null, + "software.framework.rocm": null, + "software.framework.sima": null, + "software.framework.tensorrt": "10.2", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Fujitsu", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 3.4, + "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", + "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5205 ROUGE2: 22.0796 ROUGEL: 28.6977 TOKENS_PER_SAMPLE: 292.9", + "metrics.result": 3717.74, + "metrics.result_per_accelerator": 464.7175, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", + "software.framework.ai": null, + "software.framework.cuda": "12.4", + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": null, + "software.framework.rocm": null, + "software.framework.sima": null, + "software.framework.tensorrt": "10.2", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Fujitsu", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 3.4, + "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", + "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5132 ROUGE2: 22.0915 ROUGEL: 28.7055 TOKENS_PER_SAMPLE: 293.0", + "metrics.result": 3218.55, + "metrics.result_per_accelerator": 402.31875, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", + "software.framework.ai": null, + "software.framework.cuda": "12.4", + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": null, + "software.framework.rocm": null, + "software.framework.sima": null, + "software.framework.tensorrt": "10.2", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Fujitsu", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 3.4, + "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", + "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 2845.43, - "metrics.result_per_accelerator": 2845.43, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 9960.74, + "metrics.result_per_accelerator": 2490.185, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.7", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Fujitsu", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "32x 32GB DDR5", + "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 2526.03, - "metrics.result_per_accelerator": 2526.03, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0991 ROUGE2: 20.1109 ROUGEL: 29.9733 GEN_LEN: 4114114", + "metrics.result": 9602.79, + "metrics.result_per_accelerator": 2400.6975, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.7", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Fujitsu", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "32x 32GB DDR5", + "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.2956 ROUGE2: 23.2043 ROUGEL: 30.3791 TOKENS_PER_SAMPLE: 144.9 gsm8k_accuracy: 73.0 mbxp_accuracy: 59.96", - "metrics.result": 8324.43, - "metrics.result_per_accelerator": 8324.43, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 9960.74, + "metrics.result_per_accelerator": 2490.185, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.7", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Fujitsu", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "32x 32GB DDR5", + "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.2751 ROUGE2: 23.1818 ROUGEL: 30.3401 TOKENS_PER_SAMPLE: 145.5 gsm8k_accuracy: 73.2 mbxp_accuracy: 60.1", - "metrics.result": 7735.98, - "metrics.result_per_accelerator": 7735.98, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0991 ROUGE2: 20.1109 ROUGEL: 29.9733 GEN_LEN: 4114114", + "metrics.result": 9602.79, + "metrics.result_per_accelerator": 2400.6975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.7", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Fujitsu", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "32x 32GB DDR5", + "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5446 ROUGE2: 22.1336 ROUGEL: 28.7828 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 4609.05, - "metrics.result_per_accelerator": 4609.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1787 ROUGEL: 28.7997 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 10133.3, + "metrics.result_per_accelerator": 2533.325, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.7", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Fujitsu", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "32x 32GB DDR5", + "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5531 ROUGE2: 22.1392 ROUGEL: 28.7903 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 4327.68, - "metrics.result_per_accelerator": 4327.68, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6146 ROUGE2: 22.1791 ROUGEL: 28.8001 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 9493.99, + "metrics.result_per_accelerator": 2373.4975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.7", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Fujitsu", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "32x 32GB DDR5", + "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1295 ROUGE2: 20.1909 ROUGEL: 30.052 GEN_LEN: 4153653", - "metrics.result": 6823.91, - "metrics.result_per_accelerator": 852.98875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1787 ROUGEL: 28.7997 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 10133.3, + "metrics.result_per_accelerator": 2533.325, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Fujitsu", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", - "system.cpu.core_count": 96, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.4, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", - "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", + "system.memory.configuration": "32x 32GB DDR5", + "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1396 ROUGE2: 20.1904 ROUGEL: 30.0525 GEN_LEN: 4153772", - "metrics.result": 6821.11, - "metrics.result_per_accelerator": 852.63875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6146 ROUGE2: 22.1791 ROUGEL: 28.8001 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 9493.99, + "metrics.result_per_accelerator": 2373.4975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Fujitsu", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", - "system.cpu.core_count": 96, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.4, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", - "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", + "system.memory.configuration": "32x 32GB DDR5", + "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1295 ROUGE2: 20.1909 ROUGEL: 30.052 GEN_LEN: 4153653", - "metrics.result": 6823.91, - "metrics.result_per_accelerator": 852.98875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3504 ROUGE2: 23.2436 ROUGEL: 30.3147 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.08 mbxp_accuracy: 59.9", + "metrics.result": 26641.6, + "metrics.result_per_accelerator": 6660.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Fujitsu", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", - "system.cpu.core_count": 96, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.4, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", - "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", + "system.memory.configuration": "32x 32GB DDR5", + "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1396 ROUGE2: 20.1904 ROUGEL: 30.0525 GEN_LEN: 4153772", - "metrics.result": 6821.11, - "metrics.result_per_accelerator": 852.63875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3591 ROUGE2: 23.2517 ROUGEL: 30.3173 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", + "metrics.result": 25358.5, + "metrics.result_per_accelerator": 6339.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Fujitsu", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", - "system.cpu.core_count": 96, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.4, - "system.cpu.model": "Intel(R) Xeon(R) 6740E", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", - "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", + "system.memory.configuration": "32x 32GB DDR5", + "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0194 ROUGE2: 20.0851 ROUGEL: 29.9991 GEN_LEN: 4155554", - "metrics.result": 7240.14, - "metrics.result_per_accelerator": 1810.035, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 20041.3, + "metrics.result_per_accelerator": 2505.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 9 MiB (192 instances), L1i: 6 MiB (192 instances), L2: 384 MiB (192 instances), L3: 420 MiB (4 instances)", - "system.cpu.core_count": 192, - "system.cpu.count_per_node": 4, - "system.cpu.frequency": 3.8, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8468", + "system.cpu.caches": null, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 8192.0, - "system.memory.configuration": "1TB total, 32x 32GB", - "system.name": "HPE Compute Scale-up Server 3200 (4x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "GIGABYTE G593-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0194 ROUGE2: 20.0851 ROUGEL: 29.9991 GEN_LEN: 4155554", - "metrics.result": 7179.86, - "metrics.result_per_accelerator": 1794.965, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1004 ROUGE2: 20.1119 ROUGEL: 29.9737 GEN_LEN: 4114146", + "metrics.result": 19250.4, + "metrics.result_per_accelerator": 2406.3, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "GigaComputing", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 9 MiB (192 instances), L1i: 6 MiB (192 instances), L2: 384 MiB (192 instances), L3: 420 MiB (4 instances)", - "system.cpu.core_count": 192, - "system.cpu.count_per_node": 4, - "system.cpu.frequency": 3.8, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8468", + "system.cpu.caches": null, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 8192.0, - "system.memory.configuration": "1TB total, 32x 32GB", - "system.name": "HPE Compute Scale-up Server 3200 (4x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "GIGABYTE G593-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5453 ROUGE2: 22.1331 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.6", - "metrics.result": 33781.8, - "metrics.result_per_accelerator": 4222.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 20041.3, + "metrics.result_per_accelerator": 2505.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", + "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "GIGABYTE G593-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5663 ROUGE2: 22.1467 ROUGEL: 28.7958 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 31281.1, - "metrics.result_per_accelerator": 3910.1375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1004 ROUGE2: 20.1119 ROUGEL: 29.9737 GEN_LEN: 4114146", + "metrics.result": 19250.4, + "metrics.result_per_accelerator": 2406.3, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "GigaComputing", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", + "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "GIGABYTE G593-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21014.6, - "metrics.result_per_accelerator": 2626.825, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5884 ROUGE2: 22.1723 ROUGEL: 28.7905 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 31263.8, + "metrics.result_per_accelerator": 3907.975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", + "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "GIGABYTE G593-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21019.9, - "metrics.result_per_accelerator": 2627.4875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5891 ROUGE2: 22.1726 ROUGEL: 28.7906 TOKENS_PER_SAMPLE: 291.6", + "metrics.result": 29715.3, + "metrics.result_per_accelerator": 3714.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "GigaComputing", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", + "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "GIGABYTE G593-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21324.9, - "metrics.result_per_accelerator": 2665.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5884 ROUGE2: 22.1723 ROUGEL: 28.7905 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 31263.8, + "metrics.result_per_accelerator": 3907.975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", + "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "GIGABYTE G593-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21025.5, - "metrics.result_per_accelerator": 2628.1875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5891 ROUGE2: 22.1726 ROUGEL: 28.7906 TOKENS_PER_SAMPLE: 291.6", + "metrics.result": 29715.3, + "metrics.result_per_accelerator": 3714.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "GigaComputing", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", + "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "GIGABYTE G593-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34796.8, - "metrics.result_per_accelerator": 4349.6, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3568 ROUGE2: 23.2496 ROUGEL: 30.3171 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", + "metrics.result": 59334.8, + "metrics.result_per_accelerator": 7416.85, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", + "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "GIGABYTE G593-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.497 ROUGE2: 22.0948 ROUGEL: 28.7299 TOKENS_PER_SAMPLE: 292.2", - "metrics.result": 19405.7, - "metrics.result_per_accelerator": 2425.7125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3551 ROUGE2: 23.245 ROUGEL: 30.3156 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", + "metrics.result": 57174.1, + "metrics.result_per_accelerator": 7146.7625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "GigaComputing", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", + "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "GIGABYTE G593-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34796.8, - "metrics.result_per_accelerator": 4349.6, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6156 ROUGE2: 22.1792 ROUGEL: 28.8002 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 24133.7, + "metrics.result_per_accelerator": 3016.7125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Google", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": 2.7, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8481C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.497 ROUGE2: 22.0948 ROUGEL: 28.7299 TOKENS_PER_SAMPLE: 292.2", - "metrics.result": 19405.7, - "metrics.result_per_accelerator": 2425.7125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6146 ROUGE2: 22.1791 ROUGEL: 28.7999 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 21588.3, + "metrics.result_per_accelerator": 2698.5375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Google", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": 2.7, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8481C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5453 ROUGE2: 22.133 ROUGEL: 28.7833 TOKENS_PER_SAMPLE: 291.6", - "metrics.result": 33964.0, - "metrics.result_per_accelerator": 4245.5, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6156 ROUGE2: 22.1792 ROUGEL: 28.8002 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 24133.7, + "metrics.result_per_accelerator": 3016.7125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Google", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": 2.7, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8481C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5588 ROUGE2: 22.1412 ROUGEL: 28.7922 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 31283.3, - "metrics.result_per_accelerator": 3910.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6146 ROUGE2: 22.1791 ROUGEL: 28.7999 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 21588.3, + "metrics.result_per_accelerator": 2698.5375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "RHEL 9.4", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Google", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": 2.7, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8481C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", - "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5484 ROUGE2: 22.1341 ROUGEL: 28.7828 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 9362.85, - "metrics.result_per_accelerator": 4681.425, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3671 ROUGE2: 23.2507 ROUGEL: 30.3151 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", + "metrics.result": 52524.4, + "metrics.result_per_accelerator": 6565.55, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Google", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.frequency": 2.7, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8481C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5503 ROUGE2: 22.1345 ROUGEL: 28.7855 TOKENS_PER_SAMPLE: 292.5", - "metrics.result": 8674.58, - "metrics.result_per_accelerator": 4337.29, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3561 ROUGE2: 23.2477 ROUGEL: 30.3155 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.08 mbxp_accuracy: 59.88", + "metrics.result": 50798.8, + "metrics.result_per_accelerator": 6349.85, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Google", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.frequency": 2.7, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8481C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 5666.16, - "metrics.result_per_accelerator": 2833.08, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 2632.76, + "metrics.result_per_accelerator": 2632.76, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.2", + "submission.availability": "preview", "submission.division": "closed", "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, + "system.accelerator.count_per_node": 1, "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 2, + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, "system.cpu.vendor": "NVIDIA", "system.interconnect.accelerator": "InfiniBand", "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, + "system.memory.capacity": 480.0, "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 5039.82, - "metrics.result_per_accelerator": 2519.91, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1002 ROUGE2: 20.1114 ROUGEL: 29.9742 GEN_LEN: 4114259", + "metrics.result": 2512.89, + "metrics.result_per_accelerator": 2512.89, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.2", + "submission.availability": "preview", "submission.division": "closed", "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, + "system.accelerator.count_per_node": 1, "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 2, + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, "system.cpu.vendor": "NVIDIA", "system.interconnect.accelerator": "InfiniBand", "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, + "system.memory.capacity": 480.0, "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 5666.16, - "metrics.result_per_accelerator": 2833.08, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 2632.76, + "metrics.result_per_accelerator": 2632.76, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.2", + "submission.availability": "preview", "submission.division": "closed", "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, + "system.accelerator.count_per_node": 1, "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 2, + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, "system.cpu.vendor": "NVIDIA", "system.interconnect.accelerator": "InfiniBand", "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, + "system.memory.capacity": 480.0, "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 5039.82, - "metrics.result_per_accelerator": 2519.91, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1002 ROUGE2: 20.1114 ROUGEL: 29.9742 GEN_LEN: 4114259", + "metrics.result": 2512.89, + "metrics.result_per_accelerator": 2512.89, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.2", + "submission.availability": "preview", "submission.division": "closed", "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, + "system.accelerator.count_per_node": 1, "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 2, + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, "system.cpu.vendor": "NVIDIA", "system.interconnect.accelerator": "InfiniBand", "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, + "system.memory.capacity": 480.0, "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4581 ROUGE2: 23.2422 ROUGEL: 30.2949 TOKENS_PER_SAMPLE: 145.4 gsm8k_accuracy: 73.22 mbxp_accuracy: 59.88", - "metrics.result": 16703.4, - "metrics.result_per_accelerator": 8351.7, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 4084.3, + "metrics.result_per_accelerator": 4084.3, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.2", + "submission.availability": "preview", "submission.division": "closed", "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, + "system.accelerator.count_per_node": 1, "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 2, + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, "system.cpu.vendor": "NVIDIA", "system.interconnect.accelerator": "InfiniBand", "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, + "system.memory.capacity": 480.0, "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.385 ROUGE2: 23.2684 ROUGEL: 30.2469 TOKENS_PER_SAMPLE: 146.2 gsm8k_accuracy: 72.94 mbxp_accuracy: 59.8", - "metrics.result": 15570.8, - "metrics.result_per_accelerator": 7785.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1577 ROUGEL: 28.7741 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 3884.16, + "metrics.result_per_accelerator": 3884.16, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.2", + "submission.availability": "preview", "submission.division": "closed", "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, + "system.accelerator.count_per_node": 1, "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 2, + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, "system.cpu.vendor": "NVIDIA", "system.interconnect.accelerator": "InfiniBand", "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, + "system.memory.capacity": 480.0, "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5484 ROUGE2: 22.1341 ROUGEL: 28.7828 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 9362.85, - "metrics.result_per_accelerator": 4681.425, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 4084.3, + "metrics.result_per_accelerator": 4084.3, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.2", + "submission.availability": "preview", "submission.division": "closed", "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, + "system.accelerator.count_per_node": 1, "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 2, + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, "system.cpu.vendor": "NVIDIA", "system.interconnect.accelerator": "InfiniBand", "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, + "system.memory.capacity": 480.0, "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5503 ROUGE2: 22.1345 ROUGEL: 28.7855 TOKENS_PER_SAMPLE: 292.5", - "metrics.result": 8674.58, - "metrics.result_per_accelerator": 4337.29, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1577 ROUGEL: 28.7741 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 3884.16, + "metrics.result_per_accelerator": 3884.16, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1 LTS", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.2", + "submission.availability": "preview", "submission.division": "closed", "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, + "system.accelerator.count_per_node": 1, "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 2, + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, "system.cpu.vendor": "NVIDIA", "system.interconnect.accelerator": "InfiniBand", "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, + "system.memory.capacity": 480.0, "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", - "system.number_of_nodes": 1, - "system.type": "datacenter" - }, - { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34264.0, - "metrics.result_per_accelerator": 4283.0, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", - "software.framework.ai": null, - "software.framework.cuda": "12.8", - "software.framework.cudnn": null, - "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB RAM", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", - "system.number_of_nodes": 1, - "system.type": "datacenter" - }, - { - "metrics.accuracy": "ROUGE1: 44.5513 ROUGE2: 22.1359 ROUGEL: 28.7885 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 33027.3, - "metrics.result_per_accelerator": 4128.4125, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", - "software.framework.ai": null, - "software.framework.cuda": "12.8", - "software.framework.cudnn": null, - "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB RAM", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20999.6, - "metrics.result_per_accelerator": 2624.95, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3565 ROUGE2: 23.2488 ROUGEL: 30.3157 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", + "metrics.result": 8063.02, + "metrics.result_per_accelerator": 8063.02, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.2", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB RAM", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21359.6, - "metrics.result_per_accelerator": 2669.95, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3563 ROUGE2: 23.2486 ROUGEL: 30.3162 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.94", + "metrics.result": 7450.72, + "metrics.result_per_accelerator": 7450.72, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.2", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB RAM", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20999.6, - "metrics.result_per_accelerator": 2624.95, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0343 ROUGE2: 20.0912 ROUGEL: 29.9254 GEN_LEN: 4102194", + "metrics.result": 19751.9, + "metrics.result_per_accelerator": 2468.9875, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator_host": null, "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB RAM", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21359.6, - "metrics.result_per_accelerator": 2669.95, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0345 ROUGE2: 20.0914 ROUGEL: 29.9253 GEN_LEN: 4102160", + "metrics.result": 19502.6, + "metrics.result_per_accelerator": 2437.825, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator_host": null, "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB RAM", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4064 ROUGE2: 23.1865 ROUGEL: 30.2546 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.14", - "metrics.result": 60923.3, - "metrics.result_per_accelerator": 7615.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.535 ROUGE2: 22.0991 ROUGEL: 28.7319 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 24528.3, + "metrics.result_per_accelerator": 3066.0375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator_host": null, "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB RAM", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4575 ROUGE2: 23.2383 ROUGEL: 30.3743 TOKENS_PER_SAMPLE: 145.8 gsm8k_accuracy: 73.26 mbxp_accuracy: 59.76", - "metrics.result": 57054.4, - "metrics.result_per_accelerator": 7131.8, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5353 ROUGE2: 22.0997 ROUGEL: 28.7319 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 23132.8, + "metrics.result_per_accelerator": 2891.6, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator_host": null, "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB RAM", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", - "metrics.result": 97124.0, - "metrics.result_per_accelerator": 12140.5, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.535 ROUGE2: 22.0991 ROUGEL: 28.7319 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 24424.7, + "metrics.result_per_accelerator": 3053.0875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Debian 12", - "submission.availability": "preview", + "software.operating_system": "22.04.4 LTS", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7884 ROUGE2: 22.3874 ROUGEL: 29.148 TOKENS_PER_SAMPLE: 273.8", - "metrics.result": 92338.7, - "metrics.result_per_accelerator": 11542.3375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5353 ROUGE2: 22.0997 ROUGEL: 28.732 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 23144.0, + "metrics.result_per_accelerator": 2893.0, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Debian 12", - "submission.availability": "preview", + "software.operating_system": "22.04.4 LTS", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 22.092192082812325 exact_match: 90.10742885506286 TOKENS_PER_SAMPLE: 654.1", - "metrics.result": 1498.56, - "metrics.result_per_accelerator": 187.32, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3566 ROUGE2: 23.2487 ROUGEL: 30.3174 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", + "metrics.result": 52395.1, + "metrics.result_per_accelerator": 6549.3875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Debian 12", - "submission.availability": "preview", + "software.operating_system": "22.04.4 LTS", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 22.048188322528272 exact_match: 90.08918266048973 TOKENS_PER_SAMPLE: 656.1", - "metrics.result": 847.484, - "metrics.result_per_accelerator": 105.9355, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3565 ROUGE2: 23.249 ROUGEL: 30.3137 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", + "metrics.result": 50798.5, + "metrics.result_per_accelerator": 6349.8125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Debian 12", - "submission.availability": "preview", + "software.operating_system": "22.04.4 LTS", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4539 ROUGE2: 23.217 ROUGEL: 30.2997 TOKENS_PER_SAMPLE: 144.9 gsm8k_accuracy: 73.16 mbxp_accuracy: 60.5", - "metrics.result": 121489.0, - "metrics.result_per_accelerator": 15186.125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6113 ROUGE2: 22.1451 ROUGEL: 28.7779 TOKENS_PER_SAMPLE: 290.3", + "metrics.result": 8858.78, + "metrics.result_per_accelerator": 2214.695, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Debian 12", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 22.04.4 LTS", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3 MiB, L1i: 2 MiB, L2: 128 MiB (64 instances), L3: 320 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB", + "system.name": "HPE ProLiant DL380a Gen11 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4486 ROUGE2: 23.1917 ROUGEL: 30.2808 TOKENS_PER_SAMPLE: 146.4 gsm8k_accuracy: 73.12 mbxp_accuracy: 60.4", - "metrics.result": 122411.0, - "metrics.result_per_accelerator": 15301.375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6118 ROUGE2: 22.1456 ROUGEL: 28.7778 TOKENS_PER_SAMPLE: 290.3", + "metrics.result": 6927.8, + "metrics.result_per_accelerator": 1731.95, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Debian 12", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 22.04.4 LTS", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3 MiB, L1i: 2 MiB, L2: 128 MiB (64 instances), L3: 320 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB", + "system.name": "HPE ProLiant DL380a Gen11 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", - "metrics.result": 97089.9, - "metrics.result_per_accelerator": 12136.2375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6113 ROUGE2: 22.1451 ROUGEL: 28.7779 TOKENS_PER_SAMPLE: 290.3", + "metrics.result": 8858.78, + "metrics.result_per_accelerator": 2214.695, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Debian 12", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 22.04.4 LTS", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3 MiB, L1i: 2 MiB, L2: 128 MiB (64 instances), L3: 320 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB", + "system.name": "HPE ProLiant DL380a Gen11 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7895 ROUGE2: 22.3823 ROUGEL: 29.1462 TOKENS_PER_SAMPLE: 273.8", - "metrics.result": 92203.7, - "metrics.result_per_accelerator": 11525.4625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6118 ROUGE2: 22.1456 ROUGEL: 28.7778 TOKENS_PER_SAMPLE: 290.3", + "metrics.result": 6927.8, + "metrics.result_per_accelerator": 1731.95, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Debian 12", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 22.04.4 LTS", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3 MiB, L1i: 2 MiB, L2: 128 MiB (64 instances), L3: 320 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB", + "system.name": "HPE ProLiant DL380a Gen11 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34144.7, - "metrics.result_per_accelerator": 4268.0875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6047 ROUGE2: 22.1268 ROUGEL: 28.7705 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 41672.4, + "metrics.result_per_accelerator": 2604.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "JuniperNetworks", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 16, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 112, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", - "system.cpu.vcpu_count": "224", - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", - "system.number_of_nodes": 1, + "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", + "system.name": "2x8xH100-SXM-80GB", + "system.number_of_nodes": 2, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5593 ROUGE2: 22.1417 ROUGEL: 28.7894 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 31390.2, - "metrics.result_per_accelerator": 3923.775, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6052 ROUGE2: 22.1275 ROUGEL: 28.7709 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 41091.6, + "metrics.result_per_accelerator": 2568.225, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "JuniperNetworks", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 16, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 112, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", - "system.cpu.vcpu_count": "224", - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", - "system.number_of_nodes": 1, + "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", + "system.name": "2x8xH100-SXM-80GB", + "system.number_of_nodes": 2, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21040.9, - "metrics.result_per_accelerator": 2630.1125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6047 ROUGE2: 22.1268 ROUGEL: 28.7705 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 41672.4, + "metrics.result_per_accelerator": 2604.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "JuniperNetworks", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 16, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 112, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", - "system.cpu.vcpu_count": "224", - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", - "system.number_of_nodes": 1, + "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", + "system.name": "2x8xH100-SXM-80GB", + "system.number_of_nodes": 2, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20501.8, - "metrics.result_per_accelerator": 2562.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6052 ROUGE2: 22.1275 ROUGEL: 28.7709 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 41091.6, + "metrics.result_per_accelerator": 2568.225, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "JuniperNetworks", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 16, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 112, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", - "system.cpu.vcpu_count": "224", - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", - "system.number_of_nodes": 1, + "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", + "system.name": "2x8xH100-SXM-80GB", + "system.number_of_nodes": 2, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21040.9, - "metrics.result_per_accelerator": 2630.1125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6048 ROUGE2: 22.1273 ROUGEL: 28.7697 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 82749.6, + "metrics.result_per_accelerator": 2585.925, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "JuniperNetworks", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 32, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 112, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", - "system.cpu.vcpu_count": "224", - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", - "system.number_of_nodes": 1, + "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", + "system.name": "4x8xH100-SXM-80GB", + "system.number_of_nodes": 4, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20501.8, - "metrics.result_per_accelerator": 2562.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6056 ROUGE2: 22.1274 ROUGEL: 28.7709 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 82273.2, + "metrics.result_per_accelerator": 2571.0375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "JuniperNetworks", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 32, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 112, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", - "system.cpu.vcpu_count": "224", - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", - "system.number_of_nodes": 1, + "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", + "system.name": "4x8xH100-SXM-80GB", + "system.number_of_nodes": 4, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.688728239185064 exact_match: 89.9708041032429 TOKENS_PER_SAMPLE: 645.8", - "metrics.result": 552.592, - "metrics.result_per_accelerator": 69.074, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6048 ROUGE2: 22.1273 ROUGEL: 28.7697 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 82749.6, + "metrics.result_per_accelerator": 2585.925, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "JuniperNetworks", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 32, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 112, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", - "system.cpu.vcpu_count": "224", - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", - "system.number_of_nodes": 1, + "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", + "system.name": "4x8xH100-SXM-80GB", + "system.number_of_nodes": 4, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.677865835546807 exact_match: 90.07546988749174 TOKENS_PER_SAMPLE: 648.2", - "metrics.result": 276.43, - "metrics.result_per_accelerator": 34.55375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6056 ROUGE2: 22.1274 ROUGEL: 28.7709 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 82273.2, + "metrics.result_per_accelerator": 2571.0375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 20.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "JuniperNetworks", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 32, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 112, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", - "system.cpu.vcpu_count": "224", - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", - "system.number_of_nodes": 1, + "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", + "system.name": "4x8xH100-SXM-80GB", + "system.number_of_nodes": 4, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4064 ROUGE2: 23.1865 ROUGEL: 30.2546 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.14", - "metrics.result": 61614.5, - "metrics.result_per_accelerator": 7701.8125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0773 ROUGE2: 20.0872 ROUGEL: 29.9583 GEN_LEN: 4107356", + "metrics.result": 19859.2, + "metrics.result_per_accelerator": 2482.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "Lenovo", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -6471,44 +7237,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", - "system.cpu.vcpu_count": "224", - "system.cpu.vendor": "Intel", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9454", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB HMCG94MEBRA121N", + "system.name": "ThinkSystem SR685a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3268 ROUGE2: 23.1698 ROUGEL: 30.2987 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.48 mbxp_accuracy: 59.78", - "metrics.result": 59100.1, - "metrics.result_per_accelerator": 7387.5125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0786 ROUGE2: 20.0893 ROUGEL: 29.9596 GEN_LEN: 4107309", + "metrics.result": 19716.0, + "metrics.result_per_accelerator": 2464.5, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "Lenovo", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -6517,44 +7286,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", - "system.cpu.vcpu_count": "224", - "system.cpu.vendor": "Intel", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9454", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB HMCG94MEBRA121N", + "system.name": "ThinkSystem SR685a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34144.7, - "metrics.result_per_accelerator": 4268.0875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6249 ROUGE2: 22.1876 ROUGEL: 28.8076 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 31917.0, + "metrics.result_per_accelerator": 3989.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "Lenovo", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -6563,44 +7335,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", - "system.cpu.vcpu_count": "224", - "system.cpu.vendor": "Intel", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9454", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB HMCG94MEBRA121N", + "system.name": "ThinkSystem SR685a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5593 ROUGE2: 22.1417 ROUGEL: 28.7894 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 31390.2, - "metrics.result_per_accelerator": 3923.775, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6256 ROUGE2: 22.1883 ROUGEL: 28.808 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 30068.5, + "metrics.result_per_accelerator": 3758.5625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "Lenovo", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -6609,2065 +7384,2151 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", - "system.cpu.vcpu_count": "224", - "system.cpu.vendor": "Intel", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9454", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB HMCG94MEBRA121N", + "system.name": "ThinkSystem SR685a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4701 ROUGE2: 22.0744 ROUGEL: 28.6808 TOKENS_PER_SAMPLE: 294.8", - "metrics.result": 103182.0, - "metrics.result_per_accelerator": 3224.4375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6249 ROUGE2: 22.1876 ROUGEL: 28.8076 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 31917.0, + "metrics.result_per_accelerator": 3989.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "ROCm 6.3.0", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": "6.3.0", + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS (jammy)", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "MangoBoost", + "submission.organization": "Lenovo", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", - "system.accelerator.total_count": 32, - "system.accelerator.vendor": "AMD", - "system.cpu.caches": null, - "system.cpu.core_count": 128, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9534", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 2252.8, - "system.memory.configuration": "24x 96GB Samsung M321RYGA0BB0-CQKZJ", - "system.name": "MangoBoost Mi300X (32x MI300X-192GB-HBM3, LLMBoost)", - "system.number_of_nodes": 4, - "system.type": "datacenter" - }, - { - "metrics.accuracy": "ROUGE1: 44.4557 ROUGE2: 22.0578 ROUGEL: 28.672 TOKENS_PER_SAMPLE: 295.1", - "metrics.result": 93039.9, - "metrics.result_per_accelerator": 2907.496875, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "ROCm 6.3.0", - "software.framework.ai": null, - "software.framework.cuda": null, - "software.framework.cudnn": null, - "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": "6.3.0", - "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS (jammy)", - "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "MangoBoost", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", - "system.accelerator.total_count": 32, - "system.accelerator.vendor": "AMD", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9534", + "system.cpu.model": "AMD EPYC 9454", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 2252.8, - "system.memory.configuration": "24x 96GB Samsung M321RYGA0BB0-CQKZJ", - "system.name": "MangoBoost Mi300X (32x MI300X-192GB-HBM3, LLMBoost)", - "system.number_of_nodes": 4, + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB HMCG94MEBRA121N", + "system.name": "ThinkSystem SR685a V3(8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4701 ROUGE2: 22.0744 ROUGEL: 28.6808 TOKENS_PER_SAMPLE: 294.8", - "metrics.result": 103182.0, - "metrics.result_per_accelerator": 3224.4375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6256 ROUGE2: 22.1883 ROUGEL: 28.808 TOKENS_PER_SAMPLE: 290.9", + "metrics.result": 30068.5, + "metrics.result_per_accelerator": 3758.5625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "ROCm 6.3.0", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": "6.3.0", + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS (jammy)", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "MangoBoost", - "submission.scenario": "Offline", + "submission.organization": "Lenovo", + "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", - "system.accelerator.total_count": 32, - "system.accelerator.vendor": "AMD", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9534", + "system.cpu.model": "AMD EPYC 9454", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 2252.8, - "system.memory.configuration": "24x 96GB Samsung M321RYGA0BB0-CQKZJ", - "system.name": "MangoBoost Mi300X (32x MI300X-192GB-HBM3, LLMBoost)", - "system.number_of_nodes": 4, + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB HMCG94MEBRA121N", + "system.name": "ThinkSystem SR685a V3(8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4557 ROUGE2: 22.0578 ROUGEL: 28.672 TOKENS_PER_SAMPLE: 295.1", - "metrics.result": 93039.9, - "metrics.result_per_accelerator": 2907.496875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0266 ROUGE2: 20.1037 ROUGEL: 29.9635 GEN_LEN: 4086524", + "metrics.result": 1718.33, + "metrics.result_per_accelerator": 859.165, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "ROCm 6.3.0", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.3.0, CUDA 12.3", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.3", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": "6.3.0", + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "9.3.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS (jammy)", + "software.operating_system": "Ubuntu 22.04.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "MangoBoost", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", - "system.accelerator.total_count": 32, - "system.accelerator.vendor": "AMD", + "submission.organization": "Lenovo", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 2, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 128, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 16, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9534", + "system.cpu.model": "AMD EPYC 8124P 16-Core Processor", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": null, - "system.memory.capacity": 2252.8, - "system.memory.configuration": "24x 96GB Samsung M321RYGA0BB0-CQKZJ", - "system.name": "MangoBoost Mi300X (32x MI300X-192GB-HBM3, LLMBoost)", - "system.number_of_nodes": 4, - "system.type": "datacenter" + "system.memory.capacity": 128.0, + "system.memory.configuration": "4x 32GB M321R4GA3BB6-CQKDS", + "system.name": "ThinkEdge SE455 V3 (2x NVIDIA L40S, TensorRT)", + "system.number_of_nodes": 1, + "system.type": "edge" }, { - "metrics.accuracy": "ROUGE1: 44.5042 ROUGE2: 22.0887 ROUGEL: 28.749 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 3144.18, - "metrics.result_per_accelerator": 3144.18, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.102 ROUGE2: 20.1124 ROUGEL: 29.9753 GEN_LEN: 4114427", + "metrics.result": 20552.1, + "metrics.result_per_accelerator": 2569.0125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5019 ROUGE2: 22.0906 ROUGEL: 28.7528 TOKENS_PER_SAMPLE: 292.5", - "metrics.result": 3203.87, - "metrics.result_per_accelerator": 3203.87, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.101 ROUGE2: 20.1122 ROUGEL: 29.9738 GEN_LEN: 4114231", + "metrics.result": 19785.0, + "metrics.result_per_accelerator": 2473.125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 2808.21, - "metrics.result_per_accelerator": 2808.21, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.102 ROUGE2: 20.1124 ROUGEL: 29.9753 GEN_LEN: 4114427", + "metrics.result": 20552.1, + "metrics.result_per_accelerator": 2569.0125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 2824.55, - "metrics.result_per_accelerator": 2824.55, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.101 ROUGE2: 20.1122 ROUGEL: 29.9738 GEN_LEN: 4114231", + "metrics.result": 19785.0, + "metrics.result_per_accelerator": 2473.125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 2808.21, - "metrics.result_per_accelerator": 2808.21, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5884 ROUGE2: 22.1723 ROUGEL: 28.7905 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 31973.7, + "metrics.result_per_accelerator": 3996.7125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", - "metrics.result": 2824.55, - "metrics.result_per_accelerator": 2824.55, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5866 ROUGE2: 22.1706 ROUGEL: 28.789 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 30438.7, + "metrics.result_per_accelerator": 3804.8375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4156 ROUGE2: 23.1543 ROUGEL: 30.2816 TOKENS_PER_SAMPLE: 145.2 gsm8k_accuracy: 72.94 mbxp_accuracy: 59.74", - "metrics.result": 7792.91, - "metrics.result_per_accelerator": 7792.91, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5884 ROUGE2: 22.1723 ROUGEL: 28.7905 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 31973.7, + "metrics.result_per_accelerator": 3996.7125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3856 ROUGE2: 23.246 ROUGEL: 30.2962 TOKENS_PER_SAMPLE: 146.2 gsm8k_accuracy: 73.38 mbxp_accuracy: 59.82", - "metrics.result": 7771.75, - "metrics.result_per_accelerator": 7771.75, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5866 ROUGE2: 22.1706 ROUGEL: 28.789 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 30438.7, + "metrics.result_per_accelerator": 3804.8375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5042 ROUGE2: 22.0887 ROUGEL: 28.749 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 3144.18, - "metrics.result_per_accelerator": 3144.18, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0516 ROUGE2: 20.0942 ROUGEL: 29.9288 GEN_LEN: 4111570", + "metrics.result": 2593.65, + "metrics.result_per_accelerator": 864.5500000000001, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 3, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 3, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Gold 6438N", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 512.0, + "system.memory.configuration": "16x 32GB M321R4GA3BB6-CQKDS", + "system.name": "ThinkSystem SR650 V3 (3x NVIDIA L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5019 ROUGE2: 22.0906 ROUGEL: 28.7528 TOKENS_PER_SAMPLE: 292.5", - "metrics.result": 3203.87, - "metrics.result_per_accelerator": 3203.87, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0516 ROUGE2: 20.0939 ROUGEL: 29.9288 GEN_LEN: 4111576", + "metrics.result": 2475.22, + "metrics.result_per_accelerator": 825.0733333333333, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 3, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 3, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Gold 6438N", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 512.0, + "system.memory.configuration": "16x 32GB M321R4GA3BB6-CQKDS", + "system.name": "ThinkSystem SR650 V3 (3x NVIDIA L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5767 ROUGE2: 22.1512 ROUGEL: 28.773 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 31083.6, - "metrics.result_per_accelerator": 3885.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0394 ROUGE2: 20.0943 ROUGEL: 29.9329 GEN_LEN: 4100159", + "metrics.result": 14073.6, + "metrics.result_per_accelerator": 1759.2, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 94.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 84, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9634 84-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB HMCT04MEERA131N", + "system.name": "ThinkSystem SR675 V3 (8x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5665 ROUGE2: 22.1466 ROUGEL: 28.7638 TOKENS_PER_SAMPLE: 292.9", - "metrics.result": 29916.0, - "metrics.result_per_accelerator": 3739.5, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0404 ROUGE2: 20.0949 ROUGEL: 29.9329 GEN_LEN: 4100496", + "metrics.result": 13083.5, + "metrics.result_per_accelerator": 1635.4375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 94.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 84, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9634 84-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB HMCT04MEERA131N", + "system.name": "ThinkSystem SR675 V3 (8x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20464.9, - "metrics.result_per_accelerator": 2558.1125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5999 ROUGE2: 22.1861 ROUGEL: 28.805 TOKENS_PER_SAMPLE: 291.4", + "metrics.result": 15875.8, + "metrics.result_per_accelerator": 1984.475, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 94.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 84, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9634 84-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB HMCT04MEERA131N", + "system.name": "ThinkSystem SR675 V3 (8x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 19316.9, - "metrics.result_per_accelerator": 2414.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5887 ROUGE2: 22.1721 ROUGEL: 28.7905 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 13074.7, + "metrics.result_per_accelerator": 1634.3375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 94.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 84, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9634 84-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB HMCT04MEERA131N", + "system.name": "ThinkSystem SR675 V3 (8x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20464.9, - "metrics.result_per_accelerator": 2558.1125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5999 ROUGE2: 22.1861 ROUGEL: 28.805 TOKENS_PER_SAMPLE: 291.4", + "metrics.result": 15875.8, + "metrics.result_per_accelerator": 1984.475, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 94.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 84, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9634 84-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB HMCT04MEERA131N", + "system.name": "ThinkSystem SR675 V3 (8x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 19316.9, - "metrics.result_per_accelerator": 2414.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5887 ROUGE2: 22.1721 ROUGEL: 28.7905 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 13074.7, + "metrics.result_per_accelerator": 1634.3375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.5", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.5", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 94.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 84, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9634 84-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB HMCT04MEERA131N", + "system.name": "ThinkSystem SR675 V3 (8x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.562456891920377 exact_match: 90.11248841826605 TOKENS_PER_SAMPLE: 638.5", - "metrics.result": 436.218, - "metrics.result_per_accelerator": 54.52725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.3329 ROUGE2: 22.0197 ROUGEL: 28.598 TOKENS_PER_SAMPLE: 300.5", + "metrics.result": 948.198, + "metrics.result_per_accelerator": 118.52475, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "vLLM 0.5.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.5.2", + "software.operating_system": "Ubuntu 22.04 (linux-5.15.0-94-generic-glibc2.35)", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NeuralMagic", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d cache: 1.3 MiB (40 instances), L1i cache: 1.3 MiB (40 instances), L2 cache: 40 MiB (40 instances), L3 cache: 160 MiB (5 instances)", + "system.cpu.core_count": 4, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9254 24-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": null, + "system.memory.configuration": "undefined", + "system.name": "Crusoe Cloud L40S (8x L40S PCIe, vLLM, FP8)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.565117800148553 exact_match: 90.08828921244209 TOKENS_PER_SAMPLE: 624.9", - "metrics.result": 261.98, - "metrics.result_per_accelerator": 32.7475, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.4766 ROUGE2: 22.0865 ROUGEL: 28.6953 TOKENS_PER_SAMPLE: 295.9", + "metrics.result": 592.265, + "metrics.result_per_accelerator": 74.033125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "vLLM 0.5.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.5.2", + "software.operating_system": "Ubuntu 22.04 (linux-5.15.0-94-generic-glibc2.35)", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NeuralMagic", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d cache: 1.3 MiB (40 instances), L1i cache: 1.3 MiB (40 instances), L2 cache: 40 MiB (40 instances), L3 cache: 160 MiB (5 instances)", + "system.cpu.core_count": 4, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9254 24-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": null, + "system.memory.configuration": "undefined", + "system.name": "Crusoe Cloud L40S (8x L40S PCIe, vLLM, FP8)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4914 ROUGE2: 23.3009 ROUGEL: 30.3779 TOKENS_PER_SAMPLE: 145.2 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", - "metrics.result": 51048.7, - "metrics.result_per_accelerator": 6381.0875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.7986 ROUGE2: 22.3524 ROUGEL: 29.161 TOKENS_PER_SAMPLE: 274.0", + "metrics.result": 11264.4, + "metrics.result_per_accelerator": 11264.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.1.0, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.1.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 12, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "Intel(R) Xeon(R) Silver 4410Y", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 408.0, + "system.memory.configuration": "6x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA B200 (1x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4321 ROUGE2: 23.1829 ROUGEL: 30.3742 TOKENS_PER_SAMPLE: 146.5 gsm8k_accuracy: 73.86 mbxp_accuracy: 59.88", - "metrics.result": 50924.7, - "metrics.result_per_accelerator": 6365.5875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.8023 ROUGE2: 22.355 ROUGEL: 29.1623 TOKENS_PER_SAMPLE: 273.9", + "metrics.result": 10755.6, + "metrics.result_per_accelerator": 10755.6, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.1.0, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.1.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 12, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "Intel(R) Xeon(R) Silver 4410Y", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 408.0, + "system.memory.configuration": "6x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA B200 (1x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5767 ROUGE2: 22.1512 ROUGEL: 28.773 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 31083.6, - "metrics.result_per_accelerator": 3885.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.7986 ROUGE2: 22.3524 ROUGEL: 29.161 TOKENS_PER_SAMPLE: 274.0", + "metrics.result": 11264.4, + "metrics.result_per_accelerator": 11264.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.1.0, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.1.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 12, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "Intel(R) Xeon(R) Silver 4410Y", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 408.0, + "system.memory.configuration": "6x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA B200 (1x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5665 ROUGE2: 22.1466 ROUGEL: 28.7638 TOKENS_PER_SAMPLE: 292.9", - "metrics.result": 29916.0, - "metrics.result_per_accelerator": 3739.5, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.8023 ROUGE2: 22.355 ROUGEL: 29.1623 TOKENS_PER_SAMPLE: 273.9", + "metrics.result": 10755.6, + "metrics.result_per_accelerator": 10755.6, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.1.0, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.1.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", - "submission.availability": "available", + "software.operating_system": "Ubuntu 22.04.4", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 12, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "Intel(R) Xeon(R) Silver 4410Y", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 408.0, + "system.memory.configuration": "6x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA B200 (1x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5773 ROUGE2: 22.1546 ROUGEL: 28.7713 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 7013.17, - "metrics.result_per_accelerator": 1753.2925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 19739.8, + "metrics.result_per_accelerator": 2467.475, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8.0.43", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6283 ROUGE2: 22.2205 ROUGEL: 28.8297 TOKENS_PER_SAMPLE: 290.7", - "metrics.result": 4889.92, - "metrics.result_per_accelerator": 1222.48, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0996 ROUGE2: 20.1107 ROUGEL: 29.9733 GEN_LEN: 4114082", + "metrics.result": 19233.8, + "metrics.result_per_accelerator": 2404.225, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8.0.43", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1385 ROUGE2: 20.1564 ROUGEL: 30.0269 GEN_LEN: 4155440", - "metrics.result": 6319.07, - "metrics.result_per_accelerator": 1579.7675, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 19739.8, + "metrics.result_per_accelerator": 2467.475, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8.0.43", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1385 ROUGE2: 20.1564 ROUGEL: 30.0269 GEN_LEN: 4155440", - "metrics.result": 6251.27, - "metrics.result_per_accelerator": 1562.8175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0996 ROUGE2: 20.1107 ROUGEL: 29.9733 GEN_LEN: 4114082", + "metrics.result": 19233.8, + "metrics.result_per_accelerator": 2404.225, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8.0.43", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1385 ROUGE2: 20.1564 ROUGEL: 30.0269 GEN_LEN: 4155440", - "metrics.result": 6319.07, - "metrics.result_per_accelerator": 1579.7675, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5437 ROUGE2: 22.1222 ROUGEL: 28.7396 TOKENS_PER_SAMPLE: 290.3", + "metrics.result": 24524.9, + "metrics.result_per_accelerator": 3065.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8.0.43", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1385 ROUGE2: 20.1564 ROUGEL: 30.0269 GEN_LEN: 4155440", - "metrics.result": 6251.27, - "metrics.result_per_accelerator": 1562.8175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5435 ROUGE2: 22.122 ROUGEL: 28.7395 TOKENS_PER_SAMPLE: 290.3", + "metrics.result": 21605.8, + "metrics.result_per_accelerator": 2700.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8.0.43", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3734 ROUGE2: 23.2194 ROUGEL: 30.3241 TOKENS_PER_SAMPLE: 145.4 gsm8k_accuracy: 73.22 mbxp_accuracy: 59.86", - "metrics.result": 15843.3, - "metrics.result_per_accelerator": 3960.825, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5437 ROUGE2: 22.1222 ROUGEL: 28.7396 TOKENS_PER_SAMPLE: 290.3", + "metrics.result": 24524.9, + "metrics.result_per_accelerator": 3065.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8.0.43", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3176 ROUGE2: 23.1416 ROUGEL: 30.2829 TOKENS_PER_SAMPLE: 145.4 gsm8k_accuracy: 73.52 mbxp_accuracy: 60.02", - "metrics.result": 13864.5, - "metrics.result_per_accelerator": 3466.125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5435 ROUGE2: 22.122 ROUGEL: 28.7395 TOKENS_PER_SAMPLE: 290.3", + "metrics.result": 21605.8, + "metrics.result_per_accelerator": 2700.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8.0.43", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5773 ROUGE2: 22.1546 ROUGEL: 28.7713 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 7013.17, - "metrics.result_per_accelerator": 1753.2925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3448 ROUGE2: 23.2415 ROUGEL: 30.3087 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.12 mbxp_accuracy: 59.88", + "metrics.result": 52415.9, + "metrics.result_per_accelerator": 6551.9875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8.0.43", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6283 ROUGE2: 22.2205 ROUGEL: 28.8297 TOKENS_PER_SAMPLE: 290.7", - "metrics.result": 4889.92, - "metrics.result_per_accelerator": 1222.48, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3537 ROUGE2: 23.2446 ROUGEL: 30.3114 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", + "metrics.result": 50795.6, + "metrics.result_per_accelerator": 6349.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8.0.43", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 32", - "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7466 ROUGE2: 22.3524 ROUGEL: 29.1548 TOKENS_PER_SAMPLE: 271.7", - "metrics.result": 12269.3, - "metrics.result_per_accelerator": 12269.3, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 2627.69, + "metrics.result_per_accelerator": 2627.69, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (1x B200-SXM-180GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7469 ROUGE2: 22.3525 ROUGEL: 29.1541 TOKENS_PER_SAMPLE: 272.7", - "metrics.result": 12078.4, - "metrics.result_per_accelerator": 12078.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1002 ROUGE2: 20.1114 ROUGEL: 29.9742 GEN_LEN: 4114259", + "metrics.result": 2513.38, + "metrics.result_per_accelerator": 2513.38, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (1x B200-SXM-180GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7466 ROUGE2: 22.3524 ROUGEL: 29.1548 TOKENS_PER_SAMPLE: 271.7", - "metrics.result": 12269.3, - "metrics.result_per_accelerator": 12269.3, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 2627.69, + "metrics.result_per_accelerator": 2627.69, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (1x B200-SXM-180GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7469 ROUGE2: 22.3525 ROUGEL: 29.1541 TOKENS_PER_SAMPLE: 272.7", - "metrics.result": 12078.4, - "metrics.result_per_accelerator": 12078.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1002 ROUGE2: 20.1114 ROUGEL: 29.9742 GEN_LEN: 4114259", + "metrics.result": 2513.38, + "metrics.result_per_accelerator": 2513.38, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (1x B200-SXM-180GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5775 ROUGE2: 22.1518 ROUGEL: 28.7746 TOKENS_PER_SAMPLE: 291.8", - "metrics.result": 31306.8, - "metrics.result_per_accelerator": 3913.35, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 4067.52, + "metrics.result_per_accelerator": 4067.52, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5683 ROUGE2: 22.1486 ROUGEL: 28.7643 TOKENS_PER_SAMPLE: 292.9", - "metrics.result": 31106.3, - "metrics.result_per_accelerator": 3888.2875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1576 ROUGEL: 28.7742 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 3883.67, + "metrics.result_per_accelerator": 3883.67, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5775 ROUGE2: 22.1518 ROUGEL: 28.7746 TOKENS_PER_SAMPLE: 291.8", - "metrics.result": 31306.8, - "metrics.result_per_accelerator": 3913.35, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 4067.52, + "metrics.result_per_accelerator": 4067.52, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5683 ROUGE2: 22.1486 ROUGEL: 28.7643 TOKENS_PER_SAMPLE: 292.9", - "metrics.result": 31106.3, - "metrics.result_per_accelerator": 3888.2875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1576 ROUGEL: 28.7742 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 3883.67, + "metrics.result_per_accelerator": 3883.67, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5449 ROUGE2: 22.134 ROUGEL: 28.7831 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 4686.14, - "metrics.result_per_accelerator": 4686.14, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3572 ROUGE2: 23.2487 ROUGEL: 30.3161 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.92", + "metrics.result": 8021.46, + "metrics.result_per_accelerator": 8021.46, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -8679,7 +9540,7 @@ "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, + "system.cpu.core_count": 72, "system.cpu.count_per_node": 1, "system.cpu.frequency": null, "system.cpu.model": "NVIDIA Grace CPU", @@ -8687,33 +9548,36 @@ "system.cpu.vendor": "NVIDIA", "system.interconnect.accelerator": "InfiniBand", "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, + "system.memory.capacity": 480.0, "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5529 ROUGE2: 22.1393 ROUGEL: 28.7909 TOKENS_PER_SAMPLE: 292.5", - "metrics.result": 4350.26, - "metrics.result_per_accelerator": 4350.26, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3563 ROUGE2: 23.2486 ROUGEL: 30.3161 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.94", + "metrics.result": 7450.27, + "metrics.result_per_accelerator": 7450.27, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -8725,7 +9589,7 @@ "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, + "system.cpu.core_count": 72, "system.cpu.count_per_node": 1, "system.cpu.frequency": null, "system.cpu.model": "NVIDIA Grace CPU", @@ -8733,1045 +9597,1114 @@ "system.cpu.vendor": "NVIDIA", "system.interconnect.accelerator": "InfiniBand", "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, + "system.memory.capacity": 480.0, "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 2824.83, - "metrics.result_per_accelerator": 2824.83, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 4487.88, + "metrics.result_per_accelerator": 4487.88, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", + "system.name": "NVIDIA H200 (1x H200-SXM-141GB-CTS, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 2526.75, - "metrics.result_per_accelerator": 2526.75, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5629 ROUGE2: 22.1577 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 4202.3, + "metrics.result_per_accelerator": 4202.3, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", + "system.name": "NVIDIA H200 (1x H200-SXM-141GB-CTS, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 2824.83, - "metrics.result_per_accelerator": 2824.83, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 4487.88, + "metrics.result_per_accelerator": 4487.88, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", + "system.name": "NVIDIA H200 (1x H200-SXM-141GB-CTS, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 2526.75, - "metrics.result_per_accelerator": 2526.75, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5629 ROUGE2: 22.1577 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 4202.3, + "metrics.result_per_accelerator": 4202.3, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", + "system.name": "NVIDIA H200 (1x H200-SXM-141GB-CTS, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.485 ROUGE2: 23.27 ROUGEL: 30.3119 TOKENS_PER_SAMPLE: 145.1 gsm8k_accuracy: 73.48 mbxp_accuracy: 59.94", - "metrics.result": 8345.73, - "metrics.result_per_accelerator": 8345.73, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 34864.2, + "metrics.result_per_accelerator": 4358.025, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB-CTS, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4444 ROUGE2: 23.2749 ROUGEL: 30.3007 TOKENS_PER_SAMPLE: 146.3 gsm8k_accuracy: 73.24 mbxp_accuracy: 59.88", - "metrics.result": 8042.27, - "metrics.result_per_accelerator": 8042.27, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5618 ROUGE2: 22.1563 ROUGEL: 28.7731 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 32789.7, + "metrics.result_per_accelerator": 4098.7125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB-CTS, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5449 ROUGE2: 22.134 ROUGEL: 28.7831 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 4686.14, - "metrics.result_per_accelerator": 4686.14, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 34864.2, + "metrics.result_per_accelerator": 4358.025, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB-CTS, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5529 ROUGE2: 22.1393 ROUGEL: 28.7909 TOKENS_PER_SAMPLE: 292.5", - "metrics.result": 4350.26, - "metrics.result_per_accelerator": 4350.26, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5618 ROUGE2: 22.1563 ROUGEL: 28.7731 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 32789.7, + "metrics.result_per_accelerator": 4098.7125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB-CTS, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7454 ROUGE2: 22.3503 ROUGEL: 29.1532 TOKENS_PER_SAMPLE: 271.7", - "metrics.result": 98858.0, - "metrics.result_per_accelerator": 12357.25, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 2579.51, + "metrics.result_per_accelerator": 2579.51, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (1x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7462 ROUGE2: 22.3556 ROUGEL: 29.1559 TOKENS_PER_SAMPLE: 272.6", - "metrics.result": 98443.3, - "metrics.result_per_accelerator": 12305.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1016 ROUGE2: 20.1124 ROUGEL: 29.9751 GEN_LEN: 4114376", + "metrics.result": 2405.86, + "metrics.result_per_accelerator": 2405.86, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (1x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7454 ROUGE2: 22.3503 ROUGEL: 29.1532 TOKENS_PER_SAMPLE: 271.7", - "metrics.result": 98858.0, - "metrics.result_per_accelerator": 12357.25, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 2579.51, + "metrics.result_per_accelerator": 2579.51, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (1x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.827 ROUGE2: 22.4067 ROUGEL: 29.208 TOKENS_PER_SAMPLE: 272.5", - "metrics.result": 59409.3, - "metrics.result_per_accelerator": 7426.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1016 ROUGE2: 20.1124 ROUGEL: 29.9751 GEN_LEN: 4114376", + "metrics.result": 2405.86, + "metrics.result_per_accelerator": 2405.86, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (1x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.985172700345778 exact_match: 90.12750165453342 TOKENS_PER_SAMPLE: 655.9", - "metrics.result": 1526.29, - "metrics.result_per_accelerator": 190.78625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 20086.1, + "metrics.result_per_accelerator": 2510.7625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.975590204593704 exact_match: 90.07971211118463 TOKENS_PER_SAMPLE: 654.0", - "metrics.result": 845.82, - "metrics.result_per_accelerator": 105.7275, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1027 ROUGE2: 20.1139 ROUGEL: 29.9764 GEN_LEN: 4113784", + "metrics.result": 19243.3, + "metrics.result_per_accelerator": 2405.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7454 ROUGE2: 22.3503 ROUGEL: 29.1532 TOKENS_PER_SAMPLE: 271.7", - "metrics.result": 98858.0, - "metrics.result_per_accelerator": 12357.25, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 20086.1, + "metrics.result_per_accelerator": 2510.7625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.827 ROUGE2: 22.4067 ROUGEL: 29.208 TOKENS_PER_SAMPLE: 272.5", - "metrics.result": 59409.3, - "metrics.result_per_accelerator": 7426.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1027 ROUGE2: 20.1139 ROUGEL: 29.9764 GEN_LEN: 4113784", + "metrics.result": 19243.3, + "metrics.result_per_accelerator": 2405.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.5969 ROUGE2: 23.2676 ROUGEL: 30.3668 TOKENS_PER_SAMPLE: 143.9 gsm8k_accuracy: 73.08 mbxp_accuracy: 60.3", - "metrics.result": 128148.0, - "metrics.result_per_accelerator": 16018.5, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 31302.7, + "metrics.result_per_accelerator": 3912.8375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4131 ROUGE2: 23.183 ROUGEL: 30.2308 TOKENS_PER_SAMPLE: 145.5 gsm8k_accuracy: 73.24 mbxp_accuracy: 59.7", - "metrics.result": 126845.0, - "metrics.result_per_accelerator": 15855.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5616 ROUGE2: 22.1566 ROUGEL: 28.7725 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 29228.2, + "metrics.result_per_accelerator": 3653.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7454 ROUGE2: 22.3503 ROUGEL: 29.1532 TOKENS_PER_SAMPLE: 271.7", - "metrics.result": 98858.0, - "metrics.result_per_accelerator": 12357.25, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 31302.7, + "metrics.result_per_accelerator": 3912.8375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7462 ROUGE2: 22.3556 ROUGEL: 29.1559 TOKENS_PER_SAMPLE: 272.6", - "metrics.result": 98443.3, - "metrics.result_per_accelerator": 12305.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5616 ROUGE2: 22.1566 ROUGEL: 28.7725 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 29228.2, + "metrics.result_per_accelerator": 3653.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", - "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.93672502698354 exact_match: 90.12851091992057 TOKENS_PER_SAMPLE: 656.4", - "metrics.result": 13886.4, - "metrics.result_per_accelerator": 192.86666666666667, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3568 ROUGE2: 23.2496 ROUGEL: 30.3171 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", + "metrics.result": 59022.3, + "metrics.result_per_accelerator": 7377.7875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 186.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GB200", - "system.accelerator.total_count": 72, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "for each 4-GPU node: 1.3TB LPDDR5x", - "system.name": " (72x GB200-186GB_aarch64, TensorRT, Triton)", - "system.number_of_nodes": 18, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.959911702352596 exact_match: 90.1118795499669 TOKENS_PER_SAMPLE: 656.8", - "metrics.result": 8850.1, - "metrics.result_per_accelerator": 122.91805555555555, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3535 ROUGE2: 23.2468 ROUGEL: 30.3146 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.12 mbxp_accuracy: 59.9", + "metrics.result": 57177.2, + "metrics.result_per_accelerator": 7147.15, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 186.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GB200", - "system.accelerator.total_count": 72, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 144, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "for each 4-GPU node: 1.3TB LPDDR5x", - "system.name": " (72x GB200-186GB_aarch64, TensorRT, Triton)", - "system.number_of_nodes": 18, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5388 ROUGE2: 22.127 ROUGEL: 28.7729 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 34300.2, - "metrics.result_per_accelerator": 4287.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 13096.6, + "metrics.result_per_accelerator": 1637.075, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -9793,31 +10726,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5568 ROUGE2: 22.1417 ROUGEL: 28.7946 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 33024.7, - "metrics.result_per_accelerator": 4128.0875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0987 ROUGE2: 20.1115 ROUGEL: 29.9736 GEN_LEN: 4114209", + "metrics.result": 11700.8, + "metrics.result_per_accelerator": 1462.6, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -9839,31 +10775,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5388 ROUGE2: 22.127 ROUGEL: 28.7729 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 34300.2, - "metrics.result_per_accelerator": 4287.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 13096.6, + "metrics.result_per_accelerator": 1637.075, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -9885,31 +10824,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5568 ROUGE2: 22.1417 ROUGEL: 28.7946 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 33024.7, - "metrics.result_per_accelerator": 4128.0875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0987 ROUGE2: 20.1115 ROUGEL: 29.9736 GEN_LEN: 4114209", + "metrics.result": 11700.8, + "metrics.result_per_accelerator": 1462.6, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -9931,31 +10873,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5459 ROUGE2: 22.1333 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34988.2, - "metrics.result_per_accelerator": 4373.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1578 ROUGEL: 28.7741 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 25262.1, + "metrics.result_per_accelerator": 3157.7625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -9977,31 +10922,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1489 ROUGEL: 28.797 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 33071.5, - "metrics.result_per_accelerator": 4133.9375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5622 ROUGE2: 22.1565 ROUGEL: 28.7734 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 23113.1, + "metrics.result_per_accelerator": 2889.1375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -10023,31 +10971,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21182.5, - "metrics.result_per_accelerator": 2647.8125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1578 ROUGEL: 28.7741 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 25262.1, + "metrics.result_per_accelerator": 3157.7625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -10069,31 +11020,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21544.6, - "metrics.result_per_accelerator": 2693.075, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5622 ROUGE2: 22.1565 ROUGEL: 28.7734 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 23113.1, + "metrics.result_per_accelerator": 2889.1375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -10115,31 +11069,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21182.5, - "metrics.result_per_accelerator": 2647.8125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3568 ROUGE2: 23.2496 ROUGEL: 30.3171 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", + "metrics.result": 48987.9, + "metrics.result_per_accelerator": 6123.4875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -10161,31 +11118,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21544.6, - "metrics.result_per_accelerator": 2693.075, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3548 ROUGE2: 23.2479 ROUGEL: 30.3149 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.12 mbxp_accuracy: 59.88", + "metrics.result": 45497.0, + "metrics.result_per_accelerator": 5687.125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -10207,31 +11167,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34989.2, - "metrics.result_per_accelerator": 4373.65, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5638 ROUGE2: 22.1581 ROUGEL: 28.7745 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 31059.3, + "metrics.result_per_accelerator": 3882.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -10253,31 +11216,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5031 ROUGE2: 22.1008 ROUGEL: 28.7296 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 19424.2, - "metrics.result_per_accelerator": 2428.025, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5623 ROUGE2: 22.1574 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 292.2", + "metrics.result": 30128.4, + "metrics.result_per_accelerator": 3766.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -10299,31 +11265,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.63486183033167 exact_match: 90.05081072137656 TOKENS_PER_SAMPLE: 648.6", - "metrics.result": 560.651, - "metrics.result_per_accelerator": 70.081375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5638 ROUGE2: 22.1581 ROUGEL: 28.7745 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 31059.3, + "metrics.result_per_accelerator": 3882.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -10345,31 +11314,34 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.6585777555619 exact_match: 90.07448378557245 TOKENS_PER_SAMPLE: 647.2", - "metrics.result": 291.374, - "metrics.result_per_accelerator": 36.42175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5623 ROUGE2: 22.1574 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 292.2", + "metrics.result": 30128.4, + "metrics.result_per_accelerator": 3766.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", @@ -10391,3791 +11363,4089 @@ "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34989.2, - "metrics.result_per_accelerator": 4373.65, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 42.9198 ROUGE2: 20.1178 ROUGEL: 29.9386 GEN_LEN: 4050052", + "metrics.result": 64.4734, + "metrics.result_per_accelerator": 64.4734, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "Jetpack 6.0, TensorRT 10.1, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, - "software.framework.jetpack": null, + "software.framework.jetpack": "6.0", "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.1", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Jetson r36.3.1 L4T", "submission.availability": "available", "submission.division": "closed", "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "LPDDR5", + "system.accelerator.name": "NVIDIA Jetson AGX Orin 64G", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 12, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "12-core ARM Cortex-A78AE CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "ARM", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 64.0, + "system.memory.configuration": "64GB 256-bit LPDDR5", + "system.name": "NVIDIA Jetson AGX Orin Developer Kit 64G (TensorRT)", + "system.number_of_nodes": 1, + "system.type": "edge" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", + "metrics.result": 2695.15, + "metrics.result_per_accelerator": 2695.15, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "software.framework.ai": null, + "software.framework.cuda": "12.4", + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": null, + "software.framework.rocm": null, + "software.framework.sima": null, + "software.framework.tensorrt": "10.2.0", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.2", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Oracle", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": null, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 512.0, + "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5031 ROUGE2: 22.1008 ROUGEL: 28.7296 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 19424.2, - "metrics.result_per_accelerator": 2428.025, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", + "metrics.result": 2159.58, + "metrics.result_per_accelerator": 2159.58, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Oracle", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 512.0, + "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4064 ROUGE2: 23.1865 ROUGEL: 30.2546 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.14", - "metrics.result": 62629.6, - "metrics.result_per_accelerator": 7828.7, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", + "metrics.result": 2695.15, + "metrics.result_per_accelerator": 2695.15, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Oracle", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 512.0, + "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4092 ROUGE2: 23.1612 ROUGEL: 30.3199 TOKENS_PER_SAMPLE: 144.4 gsm8k_accuracy: 73.4 mbxp_accuracy: 59.66", - "metrics.result": 61802.0, - "metrics.result_per_accelerator": 7725.25, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", + "metrics.result": 2159.58, + "metrics.result_per_accelerator": 2159.58, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Oracle", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 512.0, + "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5459 ROUGE2: 22.1333 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34988.2, - "metrics.result_per_accelerator": 4373.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0913 ROUGE2: 20.1321 ROUGEL: 29.9767 GEN_LEN: 4113606", + "metrics.result": 3463.19, + "metrics.result_per_accelerator": 865.7975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0.19", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 44, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8458P", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator": "PCIe", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 16", + "system.name": "D54U_3U_L40S_PCIe_48GBx4_TRT", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1489 ROUGEL: 28.797 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 33071.5, - "metrics.result_per_accelerator": 4133.9375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0892 ROUGE2: 20.1306 ROUGEL: 29.9753 GEN_LEN: 4113194", + "metrics.result": 3096.45, + "metrics.result_per_accelerator": 774.1125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0.19", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 44, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8458P", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator": "PCIe", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 16", + "system.name": "D54U_3U_L40S_PCIe_48GBx4_TRT", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.389 ROUGE2: 22.0197 ROUGEL: 28.6092 TOKENS_PER_SAMPLE: 297.4", - "metrics.result": 33928.1, - "metrics.result_per_accelerator": 4241.0125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0913 ROUGE2: 20.1321 ROUGEL: 29.9767 GEN_LEN: 4113606", + "metrics.result": 3463.19, + "metrics.result_per_accelerator": 865.7975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.7.0", - "software.framework.rocm": "6.3.1", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.6.5", - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.framework.tensorrt": "10.2.0.19", + "software.framework.vllm": null, + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "AMD", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 256.0, - "system.accelerator.memory_config": "HBM3E", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 4, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 44, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8458P", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2304.0, - "system.memory.configuration": "24x 96GiB Samsung M321RYGA0PB0-CWMXJ", - "system.name": "QuantaGrid D74A-7U", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "PCIe", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 16", + "system.name": "D54U_3U_L40S_PCIe_48GBx4_TRT", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4508 ROUGE2: 22.058 ROUGEL: 28.6575 TOKENS_PER_SAMPLE: 295.8", - "metrics.result": 30724.5, - "metrics.result_per_accelerator": 3840.5625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0892 ROUGE2: 20.1306 ROUGEL: 29.9753 GEN_LEN: 4113194", + "metrics.result": 3096.45, + "metrics.result_per_accelerator": 774.1125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.7.0", - "software.framework.rocm": "6.3.1", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.6.5", - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.framework.tensorrt": "10.2.0.19", + "software.framework.vllm": null, + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "AMD", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 256.0, - "system.accelerator.memory_config": "HBM3E", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 4, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 44, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8458P", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2304.0, - "system.memory.configuration": "24x 96GiB Samsung M321RYGA0PB0-CWMXJ", - "system.name": "QuantaGrid D74A-7U", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "PCIe", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 16", + "system.name": "D54U_3U_L40S_PCIe_48GBx4_TRT", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.389 ROUGE2: 22.0197 ROUGEL: 28.6092 TOKENS_PER_SAMPLE: 297.4", - "metrics.result": 33928.1, - "metrics.result_per_accelerator": 4241.0125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0658 ROUGE2: 20.1299 ROUGEL: 29.9759 GEN_LEN: 4108404", + "metrics.result": 2803.72, + "metrics.result_per_accelerator": 2803.72, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.7.0", - "software.framework.rocm": "6.3.1", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.6.5", - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.framework.tensorrt": "10.2.0.19", + "software.framework.vllm": null, + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "AMD", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 256.0, - "system.accelerator.memory_config": "HBM3E", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2304.0, - "system.memory.configuration": "24x 96GiB Samsung M321RYGA0PB0-CWMXJ", - "system.name": "QuantaGrid D74A-7U", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 512.0, + "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", + "system.name": "GH200-GraceHopper-Superchip_GH200-96GB_aarch64x1_TRT", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4508 ROUGE2: 22.058 ROUGEL: 28.6575 TOKENS_PER_SAMPLE: 295.8", - "metrics.result": 30724.5, - "metrics.result_per_accelerator": 3840.5625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0658 ROUGE2: 20.1298 ROUGEL: 29.9758 GEN_LEN: 4108405", + "metrics.result": 2160.12, + "metrics.result_per_accelerator": 2160.12, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.7.0", - "software.framework.rocm": "6.3.1", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.6.5", - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.framework.tensorrt": "10.2.0.19", + "software.framework.vllm": null, + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "AMD", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 256.0, - "system.accelerator.memory_config": "HBM3E", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2304.0, - "system.memory.configuration": "24x 96GiB Samsung M321RYGA0PB0-CWMXJ", - "system.name": "QuantaGrid D74A-7U", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 512.0, + "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", + "system.name": "GH200-GraceHopper-Superchip_GH200-96GB_aarch64x1_TRT", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34556.6, - "metrics.result_per_accelerator": 4319.575, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0658 ROUGE2: 20.1299 ROUGEL: 29.9759 GEN_LEN: 4108404", + "metrics.result": 2803.72, + "metrics.result_per_accelerator": 2803.72, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0.19", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LT", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 512.0, + "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", + "system.name": "GH200-GraceHopper-Superchip_GH200-96GB_aarch64x1_TRT", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5513 ROUGE2: 22.1381 ROUGEL: 28.7884 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 33001.0, - "metrics.result_per_accelerator": 4125.125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0658 ROUGE2: 20.1298 ROUGEL: 29.9758 GEN_LEN: 4108405", + "metrics.result": 2160.12, + "metrics.result_per_accelerator": 2160.12, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0.19", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LT", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 512.0, + "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", + "system.name": "GH200-GraceHopper-Superchip_GH200-96GB_aarch64x1_TRT", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21104.1, - "metrics.result_per_accelerator": 2638.0125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5991 ROUGE2: 22.1726 ROUGEL: 28.8022 TOKENS_PER_SAMPLE: 291.8", + "metrics.result": 3114.03, + "metrics.result_per_accelerator": 3114.03, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0.19", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LT", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 512.0, + "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", + "system.name": "GH200-GraceHopper-Superchip_GH200-96GB_aarch64x1_TRT", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21527.4, - "metrics.result_per_accelerator": 2690.925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5947 ROUGE2: 22.1631 ROUGEL: 28.793 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 2619.09, + "metrics.result_per_accelerator": 2619.09, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0.19", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LT", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 512.0, + "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", + "system.name": "GH200-GraceHopper-Superchip_GH200-96GB_aarch64x1_TRT", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21104.1, - "metrics.result_per_accelerator": 2638.0125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6053 ROUGE2: 22.2332 ROUGEL: 28.8873 TOKENS_PER_SAMPLE: 291.4", + "metrics.result": 1717.77, + "metrics.result_per_accelerator": 429.4425, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LT", + "software.operating_system": "Red Hat Enterprise Linux CoreOS release 4.14", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "RedHat", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "HBM2", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 112, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480CL", + "system.cpu.vcpu_count": "-", "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "PCIe", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB Micron DDR5", + "system.name": "L40S-RedHat-OpenShift", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21527.4, - "metrics.result_per_accelerator": 2690.925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.4633 ROUGE2: 22.0929 ROUGEL: 28.7123 TOKENS_PER_SAMPLE: 295.3", + "metrics.result": 1469.19, + "metrics.result_per_accelerator": 367.2975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LT", + "software.operating_system": "Red Hat Enterprise Linux CoreOS release 4.14", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "RedHat", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "HBM2", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 112, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480CL", + "system.cpu.vcpu_count": "-", "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "PCIe", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB Micron DDR5", + "system.name": "L40S-RedHat-OpenShift", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4064 ROUGE2: 23.1865 ROUGEL: 30.2546 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.12", - "metrics.result": 60708.6, - "metrics.result_per_accelerator": 7588.575, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 19539.2, + "metrics.result_per_accelerator": 2442.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LT", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MHz", + "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3758 ROUGE2: 23.2396 ROUGEL: 30.3046 TOKENS_PER_SAMPLE: 146.3 gsm8k_accuracy: 73.72 mbxp_accuracy: 60.0", - "metrics.result": 59149.8, - "metrics.result_per_accelerator": 7393.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1036 ROUGE2: 20.1145 ROUGEL: 29.9777 GEN_LEN: 4113966", + "metrics.result": 19810.7, + "metrics.result_per_accelerator": 2476.3375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LT", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MHz", + "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34556.6, - "metrics.result_per_accelerator": 4319.575, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 19539.2, + "metrics.result_per_accelerator": 2442.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LT", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MHz", + "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5513 ROUGE2: 22.1381 ROUGEL: 28.7884 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 33001.0, - "metrics.result_per_accelerator": 4125.125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1036 ROUGE2: 20.1145 ROUGEL: 29.9777 GEN_LEN: 4113966", + "metrics.result": 19810.7, + "metrics.result_per_accelerator": 2476.3375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LT", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MHz", + "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.3988 ROUGE2: 22.039 ROUGEL: 28.6453 TOKENS_PER_SAMPLE: 295.3", - "metrics.result": 33366.5, - "metrics.result_per_accelerator": 4170.8125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.179 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 24216.8, + "metrics.result_per_accelerator": 3027.1, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": null, - "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.7.0", - "software.framework.rocm": "6.3.1", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.6.5", - "software.operating_system": "22.04.5 LTS (Jammy Jellyfish)", + "software.framework.tensorrt": "10.2.0", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2x AMD EPYC 9475F", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2252.8, - "system.memory.configuration": "24x 96GB Samsung M321RYGA0PB2-CCPPC", - "system.name": "ASUSTeK ESC A8A-E12U (8x MI325x-256GB )", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MHz", + "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.3971 ROUGE2: 22.0537 ROUGEL: 28.6495 TOKENS_PER_SAMPLE: 295.7", - "metrics.result": 30086.5, - "metrics.result_per_accelerator": 3760.8125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1792 ROUGEL: 28.8001 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 23699.7, + "metrics.result_per_accelerator": 2962.4625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": null, - "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.7.0", - "software.framework.rocm": "6.3.1", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.6.5", - "software.operating_system": "22.04.5 LTS (Jammy Jellyfish)", + "software.framework.tensorrt": "10.2.0", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2x AMD EPYC 9475F", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2252.8, - "system.memory.configuration": "24x 96GB Samsung M321RYGA0PB2-CCPPC", - "system.name": "ASUSTeK ESC A8A-E12U (8x MI325x-256GB )", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MHz", + "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.3988 ROUGE2: 22.039 ROUGEL: 28.6453 TOKENS_PER_SAMPLE: 295.3", - "metrics.result": 33366.5, - "metrics.result_per_accelerator": 4170.8125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.179 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 24216.8, + "metrics.result_per_accelerator": 3027.1, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": null, - "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.7.0", - "software.framework.rocm": "6.3.1", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.6.5", - "software.operating_system": "22.04.5 LTS (Jammy Jellyfish)", + "software.framework.tensorrt": "10.2.0", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2x AMD EPYC 9475F", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2252.8, - "system.memory.configuration": "24x 96GB Samsung M321RYGA0PB2-CCPPC", - "system.name": "ASUSTeK ESC A8A-E12U (8x MI325x-256GB )", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MHz", + "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.3971 ROUGE2: 22.0537 ROUGEL: 28.6495 TOKENS_PER_SAMPLE: 295.7", - "metrics.result": 30086.5, - "metrics.result_per_accelerator": 3760.8125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1792 ROUGEL: 28.8001 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 23699.7, + "metrics.result_per_accelerator": 2962.4625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": null, - "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.7.0", - "software.framework.rocm": "6.3.1", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.6.5", - "software.operating_system": "22.04.5 LTS (Jammy Jellyfish)", + "software.framework.tensorrt": "10.2.0", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2x AMD EPYC 9475F", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2252.8, - "system.memory.configuration": "24x 96GB Samsung M321RYGA0PB2-CCPPC", - "system.name": "ASUSTeK ESC A8A-E12U (8x MI325x-256GB )", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MHz", + "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5429 ROUGE2: 22.1318 ROUGEL: 28.7758 TOKENS_PER_SAMPLE: 291.4", - "metrics.result": 27765.3, - "metrics.result_per_accelerator": 3470.6625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3541 ROUGE2: 23.2468 ROUGEL: 30.3101 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.16 mbxp_accuracy: 59.88", + "metrics.result": 51269.0, + "metrics.result_per_accelerator": 6408.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB ", - "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", + "system.memory.configuration": "24x 96GB DDR5 4800MHz", + "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5628 ROUGE2: 22.1477 ROUGEL: 28.7948 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 27417.3, - "metrics.result_per_accelerator": 3427.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.351 ROUGE2: 23.2445 ROUGEL: 30.3143 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", + "metrics.result": 50099.5, + "metrics.result_per_accelerator": 6262.4375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB ", - "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", + "system.memory.configuration": "24x 96GB DDR5 4800MHz", + "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", - "metrics.result": 18113.4, - "metrics.result_per_accelerator": 2264.175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 19418.9, + "metrics.result_per_accelerator": 2427.3625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", + "system.cpu.model": "AMD EPYC 9474F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB ", - "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB DDR5 4800MHz", + "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", - "metrics.result": 17953.9, - "metrics.result_per_accelerator": 2244.2375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0997 ROUGE2: 20.1116 ROUGEL: 29.9738 GEN_LEN: 4114213", + "metrics.result": 19304.4, + "metrics.result_per_accelerator": 2413.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", + "system.cpu.model": "AMD EPYC 9474F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB ", - "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB DDR5 4800MHz", + "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", - "metrics.result": 18113.4, - "metrics.result_per_accelerator": 2264.175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 19418.9, + "metrics.result_per_accelerator": 2427.3625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", + "system.cpu.model": "AMD EPYC 9474F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB ", - "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB DDR5 4800MHz", + "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", - "metrics.result": 17953.9, - "metrics.result_per_accelerator": 2244.2375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0997 ROUGE2: 20.1116 ROUGEL: 29.9738 GEN_LEN: 4114213", + "metrics.result": 19304.4, + "metrics.result_per_accelerator": 2413.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", + "system.cpu.model": "AMD EPYC 9474F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB ", - "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB DDR5 4800MHz", + "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5429 ROUGE2: 22.1318 ROUGEL: 28.7758 TOKENS_PER_SAMPLE: 291.4", - "metrics.result": 27765.3, - "metrics.result_per_accelerator": 3470.6625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 24011.7, + "metrics.result_per_accelerator": 3001.4625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", + "system.cpu.model": "AMD EPYC 9474F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB ", - "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB DDR5 4800MHz", + "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5628 ROUGE2: 22.1477 ROUGEL: 28.7948 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 27417.3, - "metrics.result_per_accelerator": 3427.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1793 ROUGEL: 28.8001 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 21775.5, + "metrics.result_per_accelerator": 2721.9375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", + "system.cpu.model": "AMD EPYC 9474F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB ", - "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB DDR5 4800MHz", + "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5468 ROUGE2: 22.134 ROUGEL: 28.7841 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34807.1, - "metrics.result_per_accelerator": 4350.8875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 24011.7, + "metrics.result_per_accelerator": 3001.4625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "AMD EPYC 9474F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB DDR5 4800MHz", + "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5621 ROUGE2: 22.1419 ROUGEL: 28.7948 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 33037.2, - "metrics.result_per_accelerator": 4129.65, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1793 ROUGEL: 28.8001 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 21775.5, + "metrics.result_per_accelerator": 2721.9375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "AMD EPYC 9474F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB DDR5 4800MHz", + "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", - "metrics.result": 21171.3, - "metrics.result_per_accelerator": 2646.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3545 ROUGE2: 23.2434 ROUGEL: 30.3185 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", + "metrics.result": 50970.6, + "metrics.result_per_accelerator": 6371.325, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "AMD EPYC 9474F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB DDR5 4800MHz", + "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", - "metrics.result": 21533.9, - "metrics.result_per_accelerator": 2691.7375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.351 ROUGE2: 23.2457 ROUGEL: 30.3126 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", + "metrics.result": 49864.0, + "metrics.result_per_accelerator": 6233.0, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "AMD EPYC 9474F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB DDR5 4800MHz", + "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", - "metrics.result": 21171.3, - "metrics.result_per_accelerator": 2646.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", + "metrics.result": 2659.47, + "metrics.result_per_accelerator": 2659.47, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", - "metrics.result": 21533.9, - "metrics.result_per_accelerator": 2691.7375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", + "metrics.result": 2159.89, + "metrics.result_per_accelerator": 2159.89, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34809.3, - "metrics.result_per_accelerator": 4351.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", + "metrics.result": 2659.47, + "metrics.result_per_accelerator": 2659.47, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5018 ROUGE2: 22.0993 ROUGEL: 28.7329 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 19418.5, - "metrics.result_per_accelerator": 2427.3125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", + "metrics.result": 2159.89, + "metrics.result_per_accelerator": 2159.89, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.697946695208735 exact_match: 90.06889808074122 TOKENS_PER_SAMPLE: 644.8", - "metrics.result": 557.097, - "metrics.result_per_accelerator": 69.637125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 19808.2, + "metrics.result_per_accelerator": 2476.025, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.678324128995275 exact_match: 90.0528821972204 TOKENS_PER_SAMPLE: 650.3", - "metrics.result": 291.183, - "metrics.result_per_accelerator": 36.397875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0997 ROUGE2: 20.1114 ROUGEL: 29.9739 GEN_LEN: 4114157", + "metrics.result": 19725.6, + "metrics.result_per_accelerator": 2465.7, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34809.3, - "metrics.result_per_accelerator": 4351.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 19808.2, + "metrics.result_per_accelerator": 2476.025, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5018 ROUGE2: 22.0993 ROUGEL: 28.7329 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 19418.5, - "metrics.result_per_accelerator": 2427.3125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0997 ROUGE2: 20.1114 ROUGEL: 29.9739 GEN_LEN: 4114157", + "metrics.result": 19725.6, + "metrics.result_per_accelerator": 2465.7, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5468 ROUGE2: 22.134 ROUGEL: 28.7841 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34807.1, - "metrics.result_per_accelerator": 4350.8875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 24180.6, + "metrics.result_per_accelerator": 3022.575, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5621 ROUGE2: 22.1419 ROUGEL: 28.7948 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 33037.2, - "metrics.result_per_accelerator": 4129.65, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1792 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 21888.6, + "metrics.result_per_accelerator": 2736.075, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB ", - "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 19936.5, - "metrics.result_per_accelerator": 2492.0625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 24180.6, + "metrics.result_per_accelerator": 3022.575, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Broadcom_Supermicro", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "Virtualized NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "32x 98GB MTC40F204WS1RC48BB1MHCC", - "system.name": " SYS-821GE-TNRT (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 19318.9, - "metrics.result_per_accelerator": 2414.8625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1792 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 21888.6, + "metrics.result_per_accelerator": 2736.075, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Broadcom_Supermicro", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "Virtualized NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "32x 98GB MTC40F204WS1RC48BB1MHCC", - "system.name": " SYS-821GE-TNRT (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 19936.5, - "metrics.result_per_accelerator": 2492.0625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3632 ROUGE2: 23.2559 ROUGEL: 30.3191 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", + "metrics.result": 52452.0, + "metrics.result_per_accelerator": 6556.5, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Broadcom_Supermicro", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "Virtualized NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "32x 98GB MTC40F204WS1RC48BB1MHCC", - "system.name": " SYS-821GE-TNRT (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 19318.9, - "metrics.result_per_accelerator": 2414.8625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3565 ROUGE2: 23.2475 ROUGEL: 30.3154 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.86", + "metrics.result": 51028.4, + "metrics.result_per_accelerator": 6378.55, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Broadcom_Supermicro", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "Virtualized NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "32x 98GB MTC40F204WS1RC48BB1MHCC", - "system.name": " SYS-821GE-TNRT (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.3988 ROUGE2: 22.039 ROUGEL: 28.6453 TOKENS_PER_SAMPLE: 295.3", - "metrics.result": 33251.3, - "metrics.result_per_accelerator": 4156.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 19803.6, + "metrics.result_per_accelerator": 2475.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "ROCm 6.3.3", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": "6.3.3", + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 2048.0, - "system.accelerator.memory_config": "HBM3E", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", - "system.cpu.caches": "512 MB", - "system.cpu.core_count": 16, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 5.0, - "system.cpu.model": "2xAMD EPY 9175F", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB Samsung M321RAJA0MB2-CCPPC", - "system.name": "G893-ZX1-AAX2", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4157 ROUGE2: 22.0162 ROUGEL: 28.6494 TOKENS_PER_SAMPLE: 296.1", - "metrics.result": 29094.3, - "metrics.result_per_accelerator": 3636.7875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1015 ROUGE2: 20.1128 ROUGEL: 29.9751 GEN_LEN: 4114215", + "metrics.result": 19635.2, + "metrics.result_per_accelerator": 2454.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "ROCm 6.3.3", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": "6.3.3", + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 2048.0, - "system.accelerator.memory_config": "HBM3E", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", - "system.cpu.caches": "512 MB", - "system.cpu.core_count": 16, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 5.0, - "system.cpu.model": "2xAMD EPY 9175F", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB Samsung M321RAJA0MB2-CCPPC", - "system.name": "G893-ZX1-AAX2", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.3988 ROUGE2: 22.039 ROUGEL: 28.6453 TOKENS_PER_SAMPLE: 295.3", - "metrics.result": 33251.3, - "metrics.result_per_accelerator": 4156.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", + "metrics.result": 19803.6, + "metrics.result_per_accelerator": 2475.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "ROCm 6.3.3", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": "6.3.3", + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 2048.0, - "system.accelerator.memory_config": "HBM3E", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", - "system.cpu.caches": "512 MB", - "system.cpu.core_count": 16, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 5.0, - "system.cpu.model": "2xAMD EPY 9175F", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB Samsung M321RAJA0MB2-CCPPC", - "system.name": "G893-ZX1-AAX2", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4157 ROUGE2: 22.0162 ROUGEL: 28.6494 TOKENS_PER_SAMPLE: 296.1", - "metrics.result": 29094.3, - "metrics.result_per_accelerator": 3636.7875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.1015 ROUGE2: 20.1128 ROUGEL: 29.9751 GEN_LEN: 4114215", + "metrics.result": 19635.2, + "metrics.result_per_accelerator": 2454.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "ROCm 6.3.3", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": "6.3.3", + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 2048.0, - "system.accelerator.memory_config": "HBM3E", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", - "system.cpu.caches": "512 MB", - "system.cpu.core_count": 16, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 5.0, - "system.cpu.model": "2xAMD EPY 9175F", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB Samsung M321RAJA0MB2-CCPPC", - "system.name": "G893-ZX1-AAX2", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34889.5, - "metrics.result_per_accelerator": 4361.1875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 24140.1, + "metrics.result_per_accelerator": 3017.5125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5553 ROUGE2: 22.1419 ROUGEL: 28.79 TOKENS_PER_SAMPLE: 292.5", - "metrics.result": 33054.0, - "metrics.result_per_accelerator": 4131.75, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6148 ROUGE2: 22.1792 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 21986.0, + "metrics.result_per_accelerator": 2748.25, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21450.8, - "metrics.result_per_accelerator": 2681.35, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 24140.1, + "metrics.result_per_accelerator": 3017.5125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21544.3, - "metrics.result_per_accelerator": 2693.0375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.6148 ROUGE2: 22.1792 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", + "metrics.result": 21986.0, + "metrics.result_per_accelerator": 2748.25, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21450.8, - "metrics.result_per_accelerator": 2681.35, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3558 ROUGE2: 23.2472 ROUGEL: 30.3111 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.16 mbxp_accuracy: 59.88", + "metrics.result": 52353.8, + "metrics.result_per_accelerator": 6544.225, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21544.3, - "metrics.result_per_accelerator": 2693.0375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3545 ROUGE2: 23.2472 ROUGEL: 30.3145 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", + "metrics.result": 50676.9, + "metrics.result_per_accelerator": 6334.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.memory.configuration": "32x 64GB DDR5 4800MHz", + "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34986.7, - "metrics.result_per_accelerator": 4373.3375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0394 ROUGE2: 20.0943 ROUGEL: 29.9329 GEN_LEN: 4100159", + "metrics.result": 19711.1, + "metrics.result_per_accelerator": 2463.8875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "InfiniBand", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5085 ROUGE2: 22.1058 ROUGEL: 28.7374 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 20234.5, - "metrics.result_per_accelerator": 2529.3125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.039 ROUGE2: 20.0945 ROUGEL: 29.933 GEN_LEN: 4100192", + "metrics.result": 19233.8, + "metrics.result_per_accelerator": 2404.225, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "InfiniBand", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.6350817800776 exact_match: 90.03330906684316 TOKENS_PER_SAMPLE: 645.1", - "metrics.result": 558.526, - "metrics.result_per_accelerator": 69.81575, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5768 ROUGE2: 22.1293 ROUGEL: 28.7645 TOKENS_PER_SAMPLE: 292.1", + "metrics.result": 24459.6, + "metrics.result_per_accelerator": 3057.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "InfiniBand", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.63593824193972 exact_match: 89.99383520847121 TOKENS_PER_SAMPLE: 644.8", - "metrics.result": 288.601, - "metrics.result_per_accelerator": 36.075125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5777 ROUGE2: 22.1298 ROUGEL: 28.7648 TOKENS_PER_SAMPLE: 292.1", + "metrics.result": 21327.0, + "metrics.result_per_accelerator": 2665.875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "InfiniBand", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34986.7, - "metrics.result_per_accelerator": 4373.3375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5768 ROUGE2: 22.1293 ROUGEL: 28.7645 TOKENS_PER_SAMPLE: 292.1", + "metrics.result": 24459.6, + "metrics.result_per_accelerator": 3057.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "InfiniBand", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5085 ROUGE2: 22.1058 ROUGEL: 28.7374 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 20234.5, - "metrics.result_per_accelerator": 2529.3125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5777 ROUGE2: 22.1298 ROUGEL: 28.7648 TOKENS_PER_SAMPLE: 292.1", + "metrics.result": 21327.0, + "metrics.result_per_accelerator": 2665.875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "InfiniBand", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4064 ROUGE2: 23.1865 ROUGEL: 30.2546 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.12", - "metrics.result": 62386.2, - "metrics.result_per_accelerator": 7798.275, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3664 ROUGE2: 23.2584 ROUGEL: 30.3279 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.08 mbxp_accuracy: 59.86", + "metrics.result": 52817.7, + "metrics.result_per_accelerator": 6602.2125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "mixtral-8x7b", "model.name": "mixtral-8x7b", "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "InfiniBand", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4078 ROUGE2: 23.2531 ROUGEL: 30.3504 TOKENS_PER_SAMPLE: 146.3 gsm8k_accuracy: 74.04 mbxp_accuracy: 59.78", - "metrics.result": 60205.0, - "metrics.result_per_accelerator": 7525.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.3528 ROUGE2: 23.2465 ROUGEL: 30.3132 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.08 mbxp_accuracy: 59.9", + "metrics.result": 50797.2, + "metrics.result_per_accelerator": 6349.65, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "mixtral-8x7b", "model.name": "mixtral-8x7b", "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "InfiniBand", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34889.5, - "metrics.result_per_accelerator": 4361.1875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0343 ROUGE2: 20.0912 ROUGEL: 29.9254 GEN_LEN: 4102192", + "metrics.result": 19889.7, + "metrics.result_per_accelerator": 2486.2125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "22.04.4 LTS", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.division": "open", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator_host": null, "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5553 ROUGE2: 22.1419 ROUGEL: 28.79 TOKENS_PER_SAMPLE: 292.5", - "metrics.result": 33054.0, - "metrics.result_per_accelerator": 4131.75, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.0346 ROUGE2: 20.0913 ROUGEL: 29.9259 GEN_LEN: 4102422", + "metrics.result": 19627.6, + "metrics.result_per_accelerator": 2453.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "22.04.4 LTS", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.division": "open", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "105 MB", - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", + "system.interconnect.accelerator_host": null, "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "G893-SD1", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 22.02784417487302 exact_match: 90.10202845797485 TOKENS_PER_SAMPLE: 655.4", - "metrics.result": 801.906, - "metrics.result_per_accelerator": 200.4765, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.4026 ROUGE2: 23.228 ROUGEL: 30.3021 TOKENS_PER_SAMPLE: 145.3 gsm8k_accuracy: 73.64 mbxp_accuracy: 59.98", + "metrics.result": 18839.0, + "metrics.result_per_accelerator": 4709.75, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "coreweave", + "submission.division": "open", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 186.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GB200", + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 144, + "system.cpu.caches": "L1d: 3 MiB, L1i: 2 MiB, L2: 128 MiB (64 instances), L3: 320 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "for each 4-GPU node: 1.3TB LPDDR5x", - "system.name": " (4 x GB200-186GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB", + "system.name": "HPE ProLiant DL380a Gen11 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.989455826099128 exact_match: 90.08294837855725 TOKENS_PER_SAMPLE: 654.2", - "metrics.result": 522.12, - "metrics.result_per_accelerator": 130.53, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 45.4098 ROUGE2: 23.2338 ROUGEL: 30.3109 TOKENS_PER_SAMPLE: 145.3 gsm8k_accuracy: 73.66 mbxp_accuracy: 60.0", + "metrics.result": 17453.4, + "metrics.result_per_accelerator": 4363.35, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 10.2.0, CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.2.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "coreweave", + "submission.division": "open", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 186.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GB200", + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 144, + "system.cpu.caches": "L1d: 3 MiB, L1i: 2 MiB, L2: 128 MiB (64 instances), L3: 320 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 960.0, - "system.memory.configuration": "for each 4-GPU node: 1.3TB LPDDR5x", - "system.name": " (4 x GB200-186GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB", + "system.name": "HPE ProLiant DL380a Gen11 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34403.5, - "metrics.result_per_accelerator": 4300.4375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.5491 ROUGE2: 22.1761 ROUGEL: 28.7877 TOKENS_PER_SAMPLE: 293.9", + "metrics.result": 1468.24, + "metrics.result_per_accelerator": 367.06, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "vLLM 0.5.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.5.2", + "software.operating_system": "Ubuntu 22.04 (linux-6.5.0-35-generic-glibc2.35)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "coreweave", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "submission.division": "open", + "submission.organization": "NeuralMagic", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d cache: 3 MiB (64 instances), L1i cache: 2 MiB (64 instances), L2 cache: 128 MiB (64 instances), L3 cache: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 4100.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": null, + "system.memory.configuration": "undefined", + "system.name": "SYS-821GE-TNHR H100 'beaker' (4x H100-SXM-80GB, vLLM, FP8)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5641 ROUGE2: 22.1454 ROUGEL: 28.794 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 33066.5, - "metrics.result_per_accelerator": 4133.3125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.2003 ROUGE2: 21.8589 ROUGEL: 28.4858 TOKENS_PER_SAMPLE: 586.9", + "metrics.result": 1577.11, + "metrics.result_per_accelerator": 394.2775, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "int4", + "software.framework": "vLLM 0.5.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.5.2", + "software.operating_system": "Ubuntu 22.04 (linux-6.5.0-35-generic-glibc2.35)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "coreweave", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "submission.division": "open", + "submission.organization": "NeuralMagic", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d cache: 3 MiB (64 instances), L1i cache: 2 MiB (64 instances), L2 cache: 128 MiB (64 instances), L3 cache: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 4100.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": null, + "system.memory.configuration": "undefined", + "system.name": "SYS-821GE-TNHR H100 'beaker' (4x H100-SXM-80GB, vLLM, GPTQ)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34403.5, - "metrics.result_per_accelerator": 4300.4375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.1657 ROUGE2: 21.8046 ROUGEL: 28.4386 TOKENS_PER_SAMPLE: 291.2", + "metrics.result": 1164.04, + "metrics.result_per_accelerator": 291.01, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "int4", + "software.framework": "vLLM 0.5.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.5.2", + "software.operating_system": "Ubuntu 22.04 (linux-6.5.0-35-generic-glibc2.35)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "coreweave", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "submission.division": "open", + "submission.organization": "NeuralMagic", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d cache: 3 MiB (64 instances), L1i cache: 2 MiB (64 instances), L2 cache: 128 MiB (64 instances), L3 cache: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 4100.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": null, + "system.memory.configuration": "undefined", + "system.name": "SYS-821GE-TNHR H100 'beaker' (4x H100-SXM-80GB, vLLM, GPTQ)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5641 ROUGE2: 22.1454 ROUGEL: 28.794 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 33066.5, - "metrics.result_per_accelerator": 4133.3125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.0047 ROUGE2: 21.7276 ROUGEL: 28.4008 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 424.895, + "metrics.result_per_accelerator": 212.4475, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp32", + "software.framework": "vLLM 0.5.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.5.2", + "software.operating_system": "Ubuntu 23.04 (linux-6.2.0-39-generic-glibc2.37)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "coreweave", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "submission.division": "open", + "submission.organization": "NeuralMagic", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 23.64703369140625, + "system.accelerator.memory_config": null, + "system.accelerator.name": "NVIDIA GeForce RTX 4090", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.caches": "L1d cache: 1.1 MiB (24 instances), L1i cache: 768 KiB (24 instances), L2 cache: 48 MiB (24 instances), L3 cache: 45 MiB (1 instance)", + "system.cpu.core_count": 24, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": 4800.0, + "system.cpu.model": "Intel(R) Xeon(R) w7-2495X", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": null, + "system.memory.capacity": null, + "system.memory.configuration": "undefined", + "system.name": "GATE Overflow Intel Sapphire Rapids RTX 4090 (2x RTX 4090, vLLM, FP8)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5825 ROUGE2: 22.1545 ROUGEL: 28.776 TOKENS_PER_SAMPLE: 291.8", - "metrics.result": 30899.7, - "metrics.result_per_accelerator": 3862.4625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.0047 ROUGE2: 21.7276 ROUGEL: 28.4008 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 424.895, + "metrics.result_per_accelerator": 212.4475, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp32", + "software.framework": "vLLM 0.5.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.5.2", + "software.operating_system": "Ubuntu 23.04 (linux-6.2.0-39-generic-glibc2.37)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "NeuralMagic", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 23.64703369140625, + "system.accelerator.memory_config": null, + "system.accelerator.name": "NVIDIA GeForce RTX 4090", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 64, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554 64-Core Processor", + "system.cpu.caches": "L1d cache: 1.1 MiB (24 instances), L1i cache: 768 KiB (24 instances), L2 cache: 48 MiB (24 instances), L3 cache: 45 MiB (1 instance)", + "system.cpu.core_count": 24, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": 4800.0, + "system.cpu.model": "Intel(R) Xeon(R) w7-2495X", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", - "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": null, + "system.memory.capacity": null, + "system.memory.configuration": "undefined", + "system.name": "GATE Overflow Intel Sapphire Rapids RTX 4090 (2x RTX 4090, vLLM, FP8)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6229 ROUGE2: 22.2108 ROUGEL: 28.8257 TOKENS_PER_SAMPLE: 291.6", - "metrics.result": 26599.4, - "metrics.result_per_accelerator": 3324.925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.1411 ROUGE2: 21.8139 ROUGEL: 28.4395 TOKENS_PER_SAMPLE: 292.9", + "metrics.result": 923.333, + "metrics.result_per_accelerator": 115.416625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp32", + "software.framework": "vLLM 0.5.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.5.2", + "software.operating_system": "Ubuntu 22.04 (linux-5.15.0-94-generic-glibc2.35)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Server", + "submission.division": "open", + "submission.organization": "NeuralMagic", + "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.caches": "L1d cache: 1.3 MiB (40 instances), L1i cache: 1.3 MiB (40 instances), L2 cache: 40 MiB (40 instances), L3 cache: 160 MiB (5 instances)", + "system.cpu.core_count": 4, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554 64-Core Processor", + "system.cpu.model": "AMD EPYC 9254 24-Core Processor", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", - "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": null, + "system.memory.configuration": "undefined", + "system.name": "Crusoe Cloud L40S (8x L40S PCIe, vLLM)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20307.7, - "metrics.result_per_accelerator": 2538.4625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.1411 ROUGE2: 21.8139 ROUGEL: 28.4395 TOKENS_PER_SAMPLE: 292.9", + "metrics.result": 923.333, + "metrics.result_per_accelerator": 115.416625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp32", + "software.framework": "vLLM 0.5.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.5.2", + "software.operating_system": "Ubuntu 22.04 (linux-5.15.0-94-generic-glibc2.35)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "NeuralMagic", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.caches": "L1d cache: 1.3 MiB (40 instances), L1i cache: 1.3 MiB (40 instances), L2 cache: 40 MiB (40 instances), L3 cache: 160 MiB (5 instances)", + "system.cpu.core_count": 4, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554 64-Core Processor", + "system.cpu.model": "AMD EPYC 9254 24-Core Processor", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", - "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": null, + "system.memory.configuration": "undefined", + "system.name": "Crusoe Cloud L40S (8x L40S PCIe, vLLM)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20706.7, - "metrics.result_per_accelerator": 2588.3375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 42.9783 ROUGE2: 24.3368 ROUGEL: 31.4041 TOKENS_PER_SAMPLE: 164.6", + "metrics.result": 11189.1, + "metrics.result_per_accelerator": 11189.1, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "submission.division": "open", + "submission.organization": "NVIDIA", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554 64-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", - "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (1x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20307.7, - "metrics.result_per_accelerator": 2538.4625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 44.46 ROUGE2: 22.4526 ROUGEL: 29.9493 TOKENS_PER_SAMPLE: 234.4", + "metrics.result": 4575.06, + "metrics.result_per_accelerator": 4575.06, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "CUDA 12.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.4", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554 64-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", - "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (1x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20706.7, - "metrics.result_per_accelerator": 2588.3375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "4.1", + "datetime": null, + "metrics.accuracy": "ROUGE1: 43.2263 ROUGE2: 24.3279 ROUGEL: 31.3816 TOKENS_PER_SAMPLE: 166.3", + "metrics.result": 184.893, + "metrics.result_per_accelerator": 184.893, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "Jetpack 6.0, TensorRT 10.1, CUDA 12.2", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.2", "software.framework.cudnn": null, - "software.framework.jetpack": null, + "software.framework.jetpack": "6.0", "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.1", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Jetson r36.3.1 L4T", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "submission.division": "open", + "submission.organization": "NVIDIA", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "LPDDR5", + "system.accelerator.name": "NVIDIA Jetson AGX Orin 64G", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 12, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554 64-Core Processor", + "system.cpu.model": "12-core ARM Cortex-A78AE CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", - "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", + "system.cpu.vendor": "ARM", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 64.0, + "system.memory.configuration": "64GB 256-bit LPDDR5", + "system.name": "NVIDIA Jetson AGX Orin Developer Kit 64G (TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.5315 ROUGE2: 23.3511 ROUGEL: 30.4188 TOKENS_PER_SAMPLE: 144.9 gsm8k_accuracy: 73.28 mbxp_accuracy: 60.06", - "metrics.result": 52724.4, - "metrics.result_per_accelerator": 6590.55, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.658590925735947 exact_match: 90.0034579748511 TOKENS_PER_SAMPLE: 648.3", + "metrics.result": 428.895, + "metrics.result_per_accelerator": 53.611875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -14187,180 +15457,192 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554 64-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", - "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", + "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.5484 ROUGE2: 23.3628 ROUGEL: 30.443 TOKENS_PER_SAMPLE: 146.5 gsm8k_accuracy: 72.96 mbxp_accuracy: 59.6", - "metrics.result": 53299.3, - "metrics.result_per_accelerator": 6662.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 30.6202 ROUGE2: 13.9221 ROUGEL: 18.9101 TOKENS_PER_SAMPLE: 581.8", + "metrics.result": 2631.93, + "metrics.result_per_accelerator": 2631.93, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "DeepSeek-R1-Distill-Llama-8B", + "model.number_of_parameters": 8.0, + "model.weight_data_types": "bfloat16", + "software.framework": "vLLM v0.7.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.5 LTS (5.15.0-131-generic)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "FlexAI", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": null, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA H100 80GB HBM3", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.caches": "L1d cache: 6.3 MiB (200 instances), L1i cache: 6.3 MiB (200 instances), L2 cache: 800 MiB (200 instances), L3 cache: 3.1 GiB (200 instances)", + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554 64-Core Processor", + "system.cpu.model": "Intel Xeon Processor (SapphireRapids)", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", - "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": null, + "system.memory.configuration": "undefined", + "system.name": "flexbench test node 0ef307db09d34a91 with 8xH100", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5825 ROUGE2: 22.1545 ROUGEL: 28.776 TOKENS_PER_SAMPLE: 291.8", - "metrics.result": 30899.7, - "metrics.result_per_accelerator": 3862.4625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.246 ROUGE2: 22.4512 ROUGEL: 27.9398 TOKENS_PER_SAMPLE: 335.5", + "metrics.result": 134.714, + "metrics.result_per_accelerator": 134.714, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "Llama-3.3-70B-Instruct-FP8-dynamic", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "bfloat16", + "software.framework": "vLLM v0.7.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.5 LTS (5.15.0-131-generic)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "submission.division": "open", + "submission.organization": "FlexAI", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": null, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA H100 80GB HBM3", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.caches": "L1d cache: 6.3 MiB (200 instances), L1i cache: 6.3 MiB (200 instances), L2 cache: 800 MiB (200 instances), L3 cache: 3.1 GiB (200 instances)", + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554 64-Core Processor", + "system.cpu.model": "Intel Xeon Processor (SapphireRapids)", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", - "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": null, + "system.memory.configuration": "undefined", + "system.name": "flexbench test node 0ef307db09d34a91 with 8xH100", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6229 ROUGE2: 22.2108 ROUGEL: 28.8257 TOKENS_PER_SAMPLE: 291.6", - "metrics.result": 26599.4, - "metrics.result_per_accelerator": 3324.925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4619 ROUGE2: 22.0574 ROUGEL: 28.6452 TOKENS_PER_SAMPLE: 295.9", + "metrics.result": 1482.31, + "metrics.result_per_accelerator": 370.5775, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "Llama-2-70b-chat-hf-FP8", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "vLLM v0.7.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.5 LTS (5.15.0-131-generic)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "FlexAI", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": null, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA H100 80GB HBM3", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.caches": "L1d cache: 6.3 MiB (200 instances), L1i cache: 6.3 MiB (200 instances), L2 cache: 800 MiB (200 instances), L3 cache: 3.1 GiB (200 instances)", + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554 64-Core Processor", + "system.cpu.model": "Intel Xeon Processor (SapphireRapids)", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", - "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe", + "system.memory.capacity": null, + "system.memory.configuration": "undefined", + "system.name": "flexbench test node 0ef307db09d34a91 with 8xH100", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0554 ROUGE2: 20.0943 ROUGEL: 29.9455 GEN_LEN: 4157784", - "metrics.result": 1754.97, - "metrics.result_per_accelerator": 877.485, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.8294 ROUGE2: 22.3814 ROUGEL: 29.1878 TOKENS_PER_SAMPLE: 271.8", + "metrics.result": 573.765, + "metrics.result_per_accelerator": 573.765, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -14371,42 +15653,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "software.operating_system": "Jetson L4T", + "submission.availability": "preview", + "submission.division": "open", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 64.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "LPDDR5X", + "system.accelerator.name": "NVIDIA Jetson AGX Thor 128G", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 14, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9224 24-Core Processor", + "system.cpu.model": "14-core ARM Poseidon-AE CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "ARM", "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB ", - "system.name": "UCS X215 M8 (2x L40S, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 128.0, + "system.memory.configuration": "128GB 256-bit LPDDR5X", + "system.name": "NVIDIA Jetson AGX Thor Developer Kit 128G (TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0249 ROUGE2: 20.0689 ROUGEL: 29.9271 GEN_LEN: 4153891", - "metrics.result": 1745.78, - "metrics.result_per_accelerator": 872.89, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.8294 ROUGE2: 22.3814 ROUGEL: 29.1878 TOKENS_PER_SAMPLE: 271.8", + "metrics.result": 573.765, + "metrics.result_per_accelerator": 573.765, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -14417,424 +15702,451 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 64.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 2, + "software.operating_system": "Jetson L4T", + "submission.availability": "preview", + "submission.division": "open", + "submission.organization": "NVIDIA", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": null, + "system.accelerator.memory_config": "LPDDR5X", + "system.accelerator.name": "NVIDIA Jetson AGX Thor 128G", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 14, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9224 24-Core Processor", + "system.cpu.model": "14-core ARM Poseidon-AE CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "ARM", "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB ", - "system.name": "UCS X215 M8 (2x L40S, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 128.0, + "system.memory.configuration": "128GB 256-bit LPDDR5X", + "system.name": "NVIDIA Jetson AGX Thor Developer Kit 128G (TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.514 ROUGE2: 22.1255 ROUGEL: 28.7527 TOKENS_PER_SAMPLE: 291.1", - "metrics.result": 3741.32, - "metrics.result_per_accelerator": 1870.66, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 47.6186 ROUGE2: 25.2204 ROUGEL: 31.4595 TOKENS_PER_SAMPLE: 296.9", + "metrics.result": 16162.8, + "metrics.result_per_accelerator": 2020.35, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama3_1-70b-fp8_dyn", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "SGLang v0.4.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "NVL", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 2, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 24, + "system.cpu.core_count": 40, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9224", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 5600 DIMMs", - "system.name": "Cisco UCS C245 M8 (2x H100NVL-PCIe-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, SGLang v0.4.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5103 ROUGE2: 22.1066 ROUGEL: 28.7387 TOKENS_PER_SAMPLE: 292.2", - "metrics.result": 3337.55, - "metrics.result_per_accelerator": 1668.775, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 18.9480793556203 exact_match: 90.14579086697552 TOKENS_PER_SAMPLE: 641.1", + "metrics.result": 243.449, + "metrics.result_per_accelerator": 30.431125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama3.1-405b", + "model.name": "deepseek-v3", + "model.number_of_parameters": 671.0, + "model.weight_data_types": "UNSET", + "software.framework": "SGLang v0.4.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "NVL", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 2, - "system.accelerator.vendor": "NVIDIA", + "submission.division": "open", + "submission.organization": "Krai", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 24, + "system.cpu.core_count": 40, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9224", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 5600 DIMMs", - "system.name": "Cisco UCS C245 M8 (2x H100NVL-PCIe-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, SGLang v0.4.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 3789.42, - "metrics.result_per_accelerator": 1894.71, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5375 ROUGE2: 22.202 ROUGEL: 28.7796 TOKENS_PER_SAMPLE: 299.2", + "metrics.result": 16229.0, + "metrics.result_per_accelerator": 2028.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-fp8_dyn", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "UNSET", + "software.framework": "vLLM v0.7.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "NVL", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 24, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9224", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 5600 DIMMs", - "system.name": "Cisco UCS C245 M8 (2x H100NVL-PCIe-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, vLLM v0.7.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 3305.27, - "metrics.result_per_accelerator": 1652.635, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6113 ROUGE2: 22.2456 ROUGEL: 28.8393 TOKENS_PER_SAMPLE: 298.1", + "metrics.result": 14426.4, + "metrics.result_per_accelerator": 1803.3, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-fp8_dyn", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "UNSET", + "software.framework": "vLLM v0.7.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "NVL", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 24, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9224", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 5600 DIMMs", - "system.name": "Cisco UCS C245 M8 (2x H100NVL-PCIe-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, vLLM v0.7.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.514 ROUGE2: 22.1255 ROUGEL: 28.7527 TOKENS_PER_SAMPLE: 291.1", - "metrics.result": 3879.95, - "metrics.result_per_accelerator": 1939.975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.59129621595158 exact_match: 90.09433818663138 TOKENS_PER_SAMPLE: 641.0", + "metrics.result": 188.604, + "metrics.result_per_accelerator": 23.5755, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp16", + "software.framework": "vLLM v0.7.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": null, - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 2, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 40, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 5600 DIMMs", - "system.name": "Cisco UCS X215c M8 with X440P PCIe node (2x H100NVL-PCIe-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, vLLM v0.7.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5111 ROUGE2: 22.1063 ROUGEL: 28.7376 TOKENS_PER_SAMPLE: 292.2", - "metrics.result": 3337.8, - "metrics.result_per_accelerator": 1668.9, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.66444823059425 exact_match: 90.10726671078757 TOKENS_PER_SAMPLE: 632.1", + "metrics.result": 278.649, + "metrics.result_per_accelerator": 34.831125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b-fp8_dyn", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp16", + "software.framework": "vLLM v0.7.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": null, - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 2, - "system.accelerator.vendor": "NVIDIA", + "submission.division": "open", + "submission.organization": "Krai", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 40, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 5600 DIMMs", - "system.name": "Cisco UCS X215c M8 with X440P PCIe node (2x H100NVL-PCIe-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, vLLM v0.7.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 3743.36, - "metrics.result_per_accelerator": 1871.68, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4974 ROUGE2: 22.1155 ROUGEL: 28.7195 TOKENS_PER_SAMPLE: 298.1", + "metrics.result": 17128.9, + "metrics.result_per_accelerator": 2141.1125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-fp8_dyn", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "vLLM v0.7.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": null, - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 2, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 40, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 5600 DIMMs", - "system.name": "Cisco UCS X215c M8 with X440P PCIe node (2x H100NVL-PCIe-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, vLLM v0.7.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 3701.78, - "metrics.result_per_accelerator": 1850.89, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5066 ROUGE2: 22.1409 ROUGEL: 28.7316 TOKENS_PER_SAMPLE: 297.1", + "metrics.result": 12989.0, + "metrics.result_per_accelerator": 1623.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-fp8_dyn", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "vLLM v0.7.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": null, - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 2, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 40, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9554", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB 5600 DIMMs", - "system.name": "Cisco UCS X215c M8 with X440P PCIe node (2x H100NVL-PCIe-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, vLLM v0.7.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34081.1, - "metrics.result_per_accelerator": 4260.1375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5183 ROUGE2: 22.1357 ROUGEL: 28.7267 TOKENS_PER_SAMPLE: 297.0", + "metrics.result": 21372.3, + "metrics.result_per_accelerator": 2671.5375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-fp8_dyn", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "vLLM v0.6.4 FlashInfer", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -14842,45 +16154,48 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", - "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, vLLM v0.6.4)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5122 ROUGE2: 22.1039 ROUGEL: 28.7358 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 30420.2, - "metrics.result_per_accelerator": 3802.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5421 ROUGE2: 22.1749 ROUGEL: 28.8061 TOKENS_PER_SAMPLE: 295.1", + "metrics.result": 19618.1, + "metrics.result_per_accelerator": 2452.2625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-fp8_dyn", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "vLLM v0.6.4 FlashInfer", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -14888,45 +16203,48 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", - "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, vLLM v0.6.4)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21557.8, - "metrics.result_per_accelerator": 2694.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5823 ROUGE2: 22.2287 ROUGEL: 28.8178 TOKENS_PER_SAMPLE: 296.5", + "metrics.result": 15906.1, + "metrics.result_per_accelerator": 1988.2625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-fp8_dyn", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "vLLM v0.6.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -14934,45 +16252,48 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", - "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, vLLM v0.6.4)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21813.4, - "metrics.result_per_accelerator": 2726.675, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5362 ROUGE2: 22.176 ROUGEL: 28.773 TOKENS_PER_SAMPLE: 295.7", + "metrics.result": 13716.7, + "metrics.result_per_accelerator": 1714.5875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-fp8_dyn", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "vLLM v0.6.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -14980,137 +16301,146 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", - "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, vLLM v0.6.4)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21557.8, - "metrics.result_per_accelerator": 2694.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4973 ROUGE2: 22.1394 ROUGEL: 28.7255 TOKENS_PER_SAMPLE: 301.3", + "metrics.result": 18561.4, + "metrics.result_per_accelerator": 2320.175, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-fp8_dyn", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "vLLM v0.6.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.core_count": 40, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", - "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, vLLM v0.6.4)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21813.4, - "metrics.result_per_accelerator": 2726.675, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5398 ROUGE2: 22.1484 ROUGEL: 28.725 TOKENS_PER_SAMPLE: 299.5", + "metrics.result": 13916.2, + "metrics.result_per_accelerator": 1739.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-fp8_dyn", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "vLLM v0.6.4", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.core_count": 40, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", - "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x MI300X-192GB, vLLM v0.6.4)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 35102.5, - "metrics.result_per_accelerator": 4387.8125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 47.5723 ROUGE2: 25.2003 ROUGEL: 31.4338 TOKENS_PER_SAMPLE: 299.6", + "metrics.result": 24624.1, + "metrics.result_per_accelerator": 3078.0125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama3_1-70b-fp8_dyn", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "SGLang v0.4.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -15118,275 +16448,293 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", - "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, SGLang v0.4.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5023 ROUGE2: 22.0975 ROUGEL: 28.7294 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 19437.4, - "metrics.result_per_accelerator": 2429.675, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 47.589 ROUGE2: 25.2681 ROUGEL: 31.545 TOKENS_PER_SAMPLE: 292.1", + "metrics.result": 27687.9, + "metrics.result_per_accelerator": 3460.9875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama3_1-70b-fp8_pre", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "SGLang v0.4.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Server", + "submission.division": "open", + "submission.organization": "Krai", + "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", - "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, SGLang v0.4.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.792712875044835 exact_match: 90.00425876902713 TOKENS_PER_SAMPLE: 660.9", - "metrics.result": 573.573, - "metrics.result_per_accelerator": 71.696625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 47.6531 ROUGE2: 25.2981 ROUGEL: 31.5596 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 26074.4, + "metrics.result_per_accelerator": 3259.3, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama3_1-70b-fp8_pre", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "SGLang v0.4.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Offline", + "submission.division": "open", + "submission.organization": "Krai", + "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", - "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, SGLang v0.4.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.698924776508495 exact_match: 90.06501323626736 TOKENS_PER_SAMPLE: 657.5", - "metrics.result": 294.275, - "metrics.result_per_accelerator": 36.784375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 47.5684 ROUGE2: 25.2712 ROUGEL: 31.533 TOKENS_PER_SAMPLE: 292.8", + "metrics.result": 27949.8, + "metrics.result_per_accelerator": 3493.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama3_1-70b-interactive-fp8_pre", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "SGLang v0.4.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Server", + "submission.division": "open", + "submission.organization": "Krai", + "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", - "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, SGLang v0.4.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34081.1, - "metrics.result_per_accelerator": 4260.1375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 47.6193 ROUGE2: 25.2818 ROUGEL: 31.5351 TOKENS_PER_SAMPLE: 291.8", + "metrics.result": 12192.4, + "metrics.result_per_accelerator": 1524.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama3_1-70b-interactive-fp8_pre", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "SGLang v0.4.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Offline", + "submission.division": "open", + "submission.organization": "Krai", + "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", - "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, SGLang v0.4.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5122 ROUGE2: 22.1039 ROUGEL: 28.7358 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 30420.2, - "metrics.result_per_accelerator": 3802.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.634266581485935 exact_match: 90.12628722700197 TOKENS_PER_SAMPLE: 656.3", + "metrics.result": 326.537, + "metrics.result_per_accelerator": 40.817125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b-fp8_dyn", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp16", + "software.framework": "SGLang v0.4.3", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Server", + "submission.division": "open", + "submission.organization": "Krai", + "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 128, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", - "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, SGLang v0.4.3)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 35385.1, - "metrics.result_per_accelerator": 4423.1375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 46.5248 ROUGE2: 24.2311 ROUGEL: 30.2319 TOKENS_PER_SAMPLE: 323.9", + "metrics.result": 30530.1, + "metrics.result_per_accelerator": 3816.2625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama3_1-70b-interactive-fp8_nim", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "NIM v1.5.0", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Lambda", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -15394,45 +16742,48 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, NIM v1.5.0)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5554 ROUGE2: 22.138 ROUGEL: 28.7885 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 33045.3, - "metrics.result_per_accelerator": 4130.6625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 46.5348 ROUGE2: 24.2605 ROUGEL: 30.246 TOKENS_PER_SAMPLE: 324.7", + "metrics.result": 15960.1, + "metrics.result_per_accelerator": 1995.0125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama3_1-70b-interactive-fp8_nim", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "NIM v1.5.0", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Lambda", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -15440,45 +16791,48 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, NIM v1.5.0)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21448.4, - "metrics.result_per_accelerator": 2681.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 46.5273 ROUGE2: 24.2307 ROUGEL: 30.2329 TOKENS_PER_SAMPLE: 323.8", + "metrics.result": 30042.8, + "metrics.result_per_accelerator": 3755.35, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama3_1-70b-fp8_nim", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "NIM v1.5.0", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Lambda", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -15486,45 +16840,48 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, NIM v1.5.0)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21543.9, - "metrics.result_per_accelerator": 2692.9875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 46.5201 ROUGE2: 24.2273 ROUGEL: 30.2319 TOKENS_PER_SAMPLE: 323.9", + "metrics.result": 28421.0, + "metrics.result_per_accelerator": 3552.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama3_1-70b-fp8_nim", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "NIM v1.5.0", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Lambda", + "submission.division": "open", + "submission.organization": "Krai", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -15532,31 +16889,34 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 (8x H200-SXM-141GB, NIM v1.5.0)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21448.4, - "metrics.result_per_accelerator": 2681.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", + "metrics.result": 97403.4, + "metrics.result_per_accelerator": 12175.425, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -15569,40 +16929,43 @@ "software.framework.vllm": null, "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "closed", + "submission.division": "open", "submission.organization": "Lambda", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21543.9, - "metrics.result_per_accelerator": 2692.9875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.6315 ROUGE2: 23.4063 ROUGEL: 30.5125 TOKENS_PER_SAMPLE: 144.1 gsm8k_accuracy: 73.06 mbxp_accuracy: 60.38", + "metrics.result": 121886.0, + "metrics.result_per_accelerator": 15235.75, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -15615,40 +16978,43 @@ "software.framework.vllm": null, "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "closed", + "submission.division": "open", "submission.organization": "Lambda", - "submission.scenario": "Server", + "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 35409.4, - "metrics.result_per_accelerator": 4426.175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", + "metrics.result": 97403.4, + "metrics.result_per_accelerator": 12175.425, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -15661,39 +17027,42 @@ "software.framework.vllm": null, "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "closed", + "submission.division": "open", "submission.organization": "Lambda", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5019 ROUGE2: 22.0977 ROUGEL: 28.7293 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 19381.0, - "metrics.result_per_accelerator": 2422.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.738071137617286 exact_match: 90.02873593646592 TOKENS_PER_SAMPLE: 660.4", + "metrics.result": 564.696, + "metrics.result_per_accelerator": 70.587, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -15707,9 +17076,9 @@ "software.framework.vllm": null, "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Lambda", - "submission.scenario": "Server", + "submission.division": "open", + "submission.organization": "Supermicro", + "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", @@ -15717,31 +17086,34 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 35409.4, - "metrics.result_per_accelerator": 4426.175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 2791.38, + "metrics.result_per_accelerator": 2791.38, "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, + "model.architecture": "LLM", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -15749,45 +17121,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Lambda", + "submission.division": "open", + "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "LPDDR5x", + "system.name": "ARS-111GL-NHR (1x GH200-96GB_aarch64)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5019 ROUGE2: 22.0977 ROUGEL: 28.7293 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 19381.0, - "metrics.result_per_accelerator": 2422.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 2791.38, + "metrics.result_per_accelerator": 2791.38, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -15795,42 +17170,45 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Lambda", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "submission.division": "open", + "submission.organization": "Supermicro", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "LPDDR5x", + "system.name": "ARS-111GL-NHR (1x GH200-96GB_aarch64)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 35385.1, - "metrics.result_per_accelerator": 4423.1375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5498 ROUGE2: 22.134 ROUGEL: 28.7842 TOKENS_PER_SAMPLE: 291.4", + "metrics.result": 8231.9, + "metrics.result_per_accelerator": 4115.95, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -15843,40 +17221,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Red Hat 9.4 + OpenShift 4.15", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Lambda", + "submission.division": "open", + "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "ARS-221GL-NHIR (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5554 ROUGE2: 22.138 ROUGEL: 28.7885 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 33045.3, - "metrics.result_per_accelerator": 4130.6625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5491 ROUGE2: 22.1346 ROUGEL: 28.7871 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 7538.8, + "metrics.result_per_accelerator": 3769.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -15889,42 +17270,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Red Hat 9.4 + OpenShift 4.15", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Lambda", + "submission.division": "open", + "submission.organization": "Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "ARS-221GL-NHIR (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", - "metrics.result": 97416.9, - "metrics.result_per_accelerator": 12177.1125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5498 ROUGE2: 22.134 ROUGEL: 28.7842 TOKENS_PER_SAMPLE: 291.4", + "metrics.result": 8231.9, + "metrics.result_per_accelerator": 4115.95, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -15935,42 +17319,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Red Hat 9.4 + OpenShift 4.15", "submission.availability": "available", - "submission.division": "closed", + "submission.division": "open", "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "ARS-221GL-NHIR (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7857 ROUGE2: 22.3844 ROUGEL: 29.1435 TOKENS_PER_SAMPLE: 273.8", - "metrics.result": 97552.1, - "metrics.result_per_accelerator": 12194.0125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5491 ROUGE2: 22.1346 ROUGEL: 28.7871 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 7538.8, + "metrics.result_per_accelerator": 3769.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -15981,42 +17368,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Red Hat 9.4 + OpenShift 4.15", "submission.availability": "available", - "submission.division": "closed", + "submission.division": "open", "submission.organization": "Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "ARS-221GL-NHIR (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.8033 ROUGE2: 22.3646 ROUGEL: 29.1684 TOKENS_PER_SAMPLE: 271.8", - "metrics.result": 92689.2, - "metrics.result_per_accelerator": 11586.15, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4322 ROUGE2: 22.0403 ROUGEL: 28.6357 TOKENS_PER_SAMPLE: 294.3", + "metrics.result": 31238.0, + "metrics.result_per_accelerator": 3904.75, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -16027,42 +17417,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.8117 ROUGE2: 22.3683 ROUGEL: 29.1731 TOKENS_PER_SAMPLE: 272.7", - "metrics.result": 62265.7, - "metrics.result_per_accelerator": 7783.2125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4297 ROUGE2: 22.0409 ROUGEL: 28.6367 TOKENS_PER_SAMPLE: 295.4", + "metrics.result": 30457.7, + "metrics.result_per_accelerator": 3807.2125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -16073,42 +17466,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 22.092192082812325 exact_match: 90.10742885506286 TOKENS_PER_SAMPLE: 654.1", - "metrics.result": 1521.74, - "metrics.result_per_accelerator": 190.2175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20796.0, + "metrics.result_per_accelerator": 2599.5, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -16119,42 +17515,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 22.071095668630047 exact_match: 90.1108107213766 TOKENS_PER_SAMPLE: 655.5", - "metrics.result": 1080.31, - "metrics.result_per_accelerator": 135.03875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 19363.3, + "metrics.result_per_accelerator": 2420.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -16165,42 +17564,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.8033 ROUGE2: 22.3646 ROUGEL: 29.1684 TOKENS_PER_SAMPLE: 271.8", - "metrics.result": 92689.2, - "metrics.result_per_accelerator": 11586.15, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20993.9, + "metrics.result_per_accelerator": 2624.2375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -16211,42 +17613,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.8117 ROUGE2: 22.3683 ROUGEL: 29.1731 TOKENS_PER_SAMPLE: 272.7", - "metrics.result": 62265.7, - "metrics.result_per_accelerator": 7783.2125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 19363.2, + "metrics.result_per_accelerator": 2420.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -16257,42 +17662,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.5759 ROUGE2: 23.2321 ROUGEL: 30.3362 TOKENS_PER_SAMPLE: 144.4 gsm8k_accuracy: 73.38 mbxp_accuracy: 60.3", - "metrics.result": 125534.0, - "metrics.result_per_accelerator": 15691.75, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4546 ROUGE2: 22.0578 ROUGEL: 28.6551 TOKENS_PER_SAMPLE: 294.3", + "metrics.result": 23389.3, + "metrics.result_per_accelerator": 2923.6625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -16303,42 +17711,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4923 ROUGE2: 23.2325 ROUGEL: 30.3051 TOKENS_PER_SAMPLE: 146.6 gsm8k_accuracy: 73.02 mbxp_accuracy: 59.72", - "metrics.result": 128961.0, - "metrics.result_per_accelerator": 16120.125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5336 ROUGE2: 22.0916 ROUGEL: 28.7043 TOKENS_PER_SAMPLE: 294.3", + "metrics.result": 14008.5, + "metrics.result_per_accelerator": 1751.0625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -16349,42 +17760,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", - "metrics.result": 97416.9, - "metrics.result_per_accelerator": 12177.1125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4546 ROUGE2: 22.0578 ROUGEL: 28.6551 TOKENS_PER_SAMPLE: 294.3", + "metrics.result": 23385.0, + "metrics.result_per_accelerator": 2923.125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -16395,42 +17809,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7857 ROUGE2: 22.3844 ROUGEL: 29.1435 TOKENS_PER_SAMPLE: 273.8", - "metrics.result": 97552.1, - "metrics.result_per_accelerator": 12194.0125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5144 ROUGE2: 22.0732 ROUGEL: 28.686 TOKENS_PER_SAMPLE: 294.4", + "metrics.result": 13987.6, + "metrics.result_per_accelerator": 1748.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -16441,40 +17858,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34628.3, - "metrics.result_per_accelerator": 4328.5375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4347 ROUGE2: 22.0435 ROUGEL: 28.6357 TOKENS_PER_SAMPLE: 294.3", + "metrics.result": 31252.1, + "metrics.result_per_accelerator": 3906.5125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -16487,40 +17907,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5615 ROUGE2: 22.1445 ROUGEL: 28.7927 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 33052.9, - "metrics.result_per_accelerator": 4131.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4184 ROUGE2: 22.033 ROUGEL: 28.6265 TOKENS_PER_SAMPLE: 295.6", + "metrics.result": 30500.8, + "metrics.result_per_accelerator": 3812.6, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -16533,41 +17956,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with HPE GreenLake for File Storage (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21304.3, - "metrics.result_per_accelerator": 2663.0375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5458 ROUGE2: 22.1332 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 30297.5, + "metrics.result_per_accelerator": 3787.1875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -16579,41 +18005,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", + "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21544.2, - "metrics.result_per_accelerator": 2693.025, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5621 ROUGE2: 22.1437 ROUGEL: 28.7928 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 27638.9, + "metrics.result_per_accelerator": 3454.8625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -16625,40 +18054,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", + "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21304.3, - "metrics.result_per_accelerator": 2663.0375, + "metrics.result": 17882.9, + "metrics.result_per_accelerator": 2235.3625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -16671,40 +18103,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", + "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21544.2, - "metrics.result_per_accelerator": 2693.025, + "metrics.result": 18065.1, + "metrics.result_per_accelerator": 2258.1375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -16717,41 +18152,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", + "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5467 ROUGE2: 22.1337 ROUGEL: 28.7837 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34667.6, - "metrics.result_per_accelerator": 4333.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 17882.9, + "metrics.result_per_accelerator": 2235.3625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -16763,41 +18201,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", + "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5006 ROUGE2: 22.1008 ROUGEL: 28.7314 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 19431.9, - "metrics.result_per_accelerator": 2428.9875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 18065.1, + "metrics.result_per_accelerator": 2258.1375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -16809,41 +18250,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", + "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5467 ROUGE2: 22.1337 ROUGEL: 28.7837 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34667.6, - "metrics.result_per_accelerator": 4333.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.3505 ROUGE2: 23.2113 ROUGEL: 30.312 TOKENS_PER_SAMPLE: 144.9 gsm8k_accuracy: 73.02 mbxp_accuracy: 59.82", + "metrics.result": 52205.9, + "metrics.result_per_accelerator": 6525.7375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -16855,41 +18299,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", + "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5006 ROUGE2: 22.1008 ROUGEL: 28.7314 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 19431.9, - "metrics.result_per_accelerator": 2428.9875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4369 ROUGE2: 23.2825 ROUGEL: 30.4229 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.48 mbxp_accuracy: 59.94", + "metrics.result": 50123.9, + "metrics.result_per_accelerator": 6265.4875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -16901,41 +18348,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", + "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4086 ROUGE2: 23.1851 ROUGEL: 30.262 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.12", - "metrics.result": 61964.1, - "metrics.result_per_accelerator": 7745.5125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5458 ROUGE2: 22.1332 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 30297.5, + "metrics.result_per_accelerator": 3787.1875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -16947,41 +18397,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", + "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4991 ROUGE2: 23.3568 ROUGEL: 30.4444 TOKENS_PER_SAMPLE: 146.2 gsm8k_accuracy: 73.64 mbxp_accuracy: 60.14", - "metrics.result": 59627.3, - "metrics.result_per_accelerator": 7453.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5621 ROUGE2: 22.1437 ROUGEL: 28.7928 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 27638.9, + "metrics.result_per_accelerator": 3454.8625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -16993,40 +18446,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "16x 128GB MTC40F2047S1RC64BB1 MWFF; 16x 64GB HMCG94AHBRA480N", + "system.name": "HPE ProLiant DL380a Compute Gen12 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34628.3, - "metrics.result_per_accelerator": 4328.5375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4331 ROUGE2: 22.0455 ROUGEL: 28.6365 TOKENS_PER_SAMPLE: 294.2", + "metrics.result": 31174.7, + "metrics.result_per_accelerator": 3896.8375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -17039,40 +18495,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5615 ROUGE2: 22.1445 ROUGEL: 28.7927 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 33052.9, - "metrics.result_per_accelerator": 4131.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4303 ROUGE2: 22.039 ROUGEL: 28.6356 TOKENS_PER_SAMPLE: 295.3", + "metrics.result": 30498.9, + "metrics.result_per_accelerator": 3812.3625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -17085,43 +18544,46 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", - "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4028 ROUGE2: 23.2139 ROUGEL: 30.3225 TOKENS_PER_SAMPLE: 145.1 gsm8k_accuracy: 73.14 mbxp_accuracy: 59.62", - "metrics.result": 7718.34, - "metrics.result_per_accelerator": 7718.34, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20716.5, + "metrics.result_per_accelerator": 2589.5625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", - "software.framework": "CUDA 12.8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -17129,45 +18591,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "LPDDR5x", - "system.name": "ARS-111GL-NHR (1x GH200-96GB_aarch64)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.5543 ROUGE2: 23.3316 ROUGEL: 30.4478 TOKENS_PER_SAMPLE: 146.1 gsm8k_accuracy: 73.72 mbxp_accuracy: 60.2", - "metrics.result": 7182.77, - "metrics.result_per_accelerator": 7182.77, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20795.3, + "metrics.result_per_accelerator": 2599.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", - "software.framework": "CUDA 12.8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -17175,43 +18640,46 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "LPDDR5x", - "system.name": "ARS-111GL-NHR (1x GH200-96GB_aarch64)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7836 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 35026.8, - "metrics.result_per_accelerator": 4378.35, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20925.3, + "metrics.result_per_accelerator": 2615.6625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -17223,41 +18691,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5599 ROUGE2: 22.1445 ROUGEL: 28.7952 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 33045.7, - "metrics.result_per_accelerator": 4130.7125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20991.3, + "metrics.result_per_accelerator": 2623.9125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -17269,41 +18740,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21477.4, - "metrics.result_per_accelerator": 2684.675, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5218 ROUGE2: 22.1082 ROUGEL: 28.731 TOKENS_PER_SAMPLE: 293.1", + "metrics.result": 23435.4, + "metrics.result_per_accelerator": 2929.425, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -17315,41 +18789,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21757.3, - "metrics.result_per_accelerator": 2719.6625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5203 ROUGE2: 22.0752 ROUGEL: 28.6917 TOKENS_PER_SAMPLE: 294.2", + "metrics.result": 14004.9, + "metrics.result_per_accelerator": 1750.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -17361,41 +18838,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21477.4, - "metrics.result_per_accelerator": 2684.675, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5218 ROUGE2: 22.1082 ROUGEL: 28.731 TOKENS_PER_SAMPLE: 293.1", + "metrics.result": 23653.7, + "metrics.result_per_accelerator": 2956.7125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -17407,41 +18887,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21757.3, - "metrics.result_per_accelerator": 2719.6625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5263 ROUGE2: 22.0802 ROUGEL: 28.6949 TOKENS_PER_SAMPLE: 294.6", + "metrics.result": 14014.1, + "metrics.result_per_accelerator": 1751.7625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -17453,40 +18936,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1334 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 35192.5, - "metrics.result_per_accelerator": 4399.0625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4346 ROUGE2: 22.0435 ROUGEL: 28.6357 TOKENS_PER_SAMPLE: 294.3", + "metrics.result": 31147.4, + "metrics.result_per_accelerator": 3893.425, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -17499,40 +18985,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5031 ROUGE2: 22.1063 ROUGEL: 28.7341 TOKENS_PER_SAMPLE: 292.5", - "metrics.result": 19413.2, - "metrics.result_per_accelerator": 2426.65, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4347 ROUGE2: 22.0423 ROUGEL: 28.6376 TOKENS_PER_SAMPLE: 295.4", + "metrics.result": 30480.6, + "metrics.result_per_accelerator": 3810.075, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -17545,43 +19034,46 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "L1d: 5.3 MiB (112 instances), L1i: 3.5 MiB (112 instances), L2: 224 MiB (112 instances), L3: 210 MiB (2 instances)", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB DDR5-4800 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.50967980348964 exact_match: 90.01451025810722 TOKENS_PER_SAMPLE: 632.9", - "metrics.result": 551.221, - "metrics.result_per_accelerator": 68.902625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5024 ROUGE2: 22.0807 ROUGEL: 28.7026 TOKENS_PER_SAMPLE: 293.0", + "metrics.result": 3655.89, + "metrics.result_per_accelerator": 456.98625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.0, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -17589,45 +19081,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, + "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 2.4, + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", + "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.654998586911255 exact_match: 90.01192256783585 TOKENS_PER_SAMPLE: 646.4", - "metrics.result": 290.093, - "metrics.result_per_accelerator": 36.261625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5208 ROUGE2: 22.0821 ROUGEL: 28.7224 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 3181.65, + "metrics.result_per_accelerator": 397.70625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.0, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -17635,45 +19130,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, + "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 2.4, + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", + "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1334 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 35192.5, - "metrics.result_per_accelerator": 4399.0625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5024 ROUGE2: 22.0807 ROUGEL: 28.7026 TOKENS_PER_SAMPLE: 293.0", + "metrics.result": 3655.89, + "metrics.result_per_accelerator": 456.98625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.0, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -17681,45 +19179,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, + "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 2.4, + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", + "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5031 ROUGE2: 22.1063 ROUGEL: 28.7341 TOKENS_PER_SAMPLE: 292.5", - "metrics.result": 19413.2, - "metrics.result_per_accelerator": 2426.65, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5208 ROUGE2: 22.0821 ROUGEL: 28.7224 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 3181.65, + "metrics.result_per_accelerator": 397.70625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.weight_data_types": "fp16", + "software.framework": "TensorRT 9.0, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -17727,47 +19228,50 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "9.0", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, + "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 2.4, + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", + "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4192 ROUGE2: 23.1932 ROUGEL: 30.2622 TOKENS_PER_SAMPLE: 145.7 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.12", - "metrics.result": 62766.2, - "metrics.result_per_accelerator": 7845.775, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5446 ROUGE2: 22.1336 ROUGEL: 28.7828 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 4609.05, + "metrics.result_per_accelerator": 4609.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, @@ -17775,45 +19279,48 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.518 ROUGE2: 23.1971 ROUGEL: 30.3919 TOKENS_PER_SAMPLE: 145.8 gsm8k_accuracy: 73.0 mbxp_accuracy: 59.7", - "metrics.result": 59647.1, - "metrics.result_per_accelerator": 7455.8875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5531 ROUGE2: 22.1392 ROUGEL: 28.7903 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 4327.68, + "metrics.result_per_accelerator": 4327.68, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, @@ -17821,45 +19328,48 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7836 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 35026.8, - "metrics.result_per_accelerator": 4378.35, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 2845.43, + "metrics.result_per_accelerator": 2845.43, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, @@ -17867,45 +19377,48 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5599 ROUGE2: 22.1445 ROUGEL: 28.7952 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 33045.7, - "metrics.result_per_accelerator": 4130.7125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 2526.03, + "metrics.result_per_accelerator": 2526.03, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, @@ -17913,45 +19426,48 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MT/s", - "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", - "metrics.result": 98782.4, - "metrics.result_per_accelerator": 12347.8, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 2845.43, + "metrics.result_per_accelerator": 2845.43, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, @@ -17959,45 +19475,48 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7848 ROUGE2: 22.3844 ROUGEL: 29.1432 TOKENS_PER_SAMPLE: 273.8", - "metrics.result": 97629.8, - "metrics.result_per_accelerator": 12203.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 2526.03, + "metrics.result_per_accelerator": 2526.03, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, @@ -18005,45 +19524,48 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.8033 ROUGE2: 22.3646 ROUGEL: 29.1684 TOKENS_PER_SAMPLE: 271.8", - "metrics.result": 93857.0, - "metrics.result_per_accelerator": 11732.125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.2956 ROUGE2: 23.2043 ROUGEL: 30.3791 TOKENS_PER_SAMPLE: 144.9 gsm8k_accuracy: 73.0 mbxp_accuracy: 59.96", + "metrics.result": 8324.43, + "metrics.result_per_accelerator": 8324.43, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, @@ -18051,45 +19573,48 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.8002 ROUGE2: 22.3621 ROUGEL: 29.1653 TOKENS_PER_SAMPLE: 272.8", - "metrics.result": 59505.3, - "metrics.result_per_accelerator": 7438.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.2751 ROUGE2: 23.1818 ROUGEL: 30.3401 TOKENS_PER_SAMPLE: 145.5 gsm8k_accuracy: 73.2 mbxp_accuracy: 60.1", + "metrics.result": 7735.98, + "metrics.result_per_accelerator": 7735.98, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, @@ -18097,45 +19622,48 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 22.092192082812325 exact_match: 90.10742885506288 TOKENS_PER_SAMPLE: 654.1", - "metrics.result": 1538.17, - "metrics.result_per_accelerator": 192.27125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5446 ROUGE2: 22.1336 ROUGEL: 28.7828 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 4609.05, + "metrics.result_per_accelerator": 4609.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, @@ -18143,45 +19671,48 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 22.044077554195237 exact_match: 90.11247518199869 TOKENS_PER_SAMPLE: 657.3", - "metrics.result": 1057.52, - "metrics.result_per_accelerator": 132.19, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5531 ROUGE2: 22.1392 ROUGEL: 28.7903 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 4327.68, + "metrics.result_per_accelerator": 4327.68, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp4", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.7", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": "12.7", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, @@ -18189,42 +19720,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.8033 ROUGE2: 22.3646 ROUGEL: 29.1684 TOKENS_PER_SAMPLE: 271.8", - "metrics.result": 93857.0, - "metrics.result_per_accelerator": 11732.125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1295 ROUGE2: 20.1909 ROUGEL: 30.052 GEN_LEN: 4153653", + "metrics.result": 6823.91, + "metrics.result_per_accelerator": 852.98875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -18235,42 +19769,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": 2.4, + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", + "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.8002 ROUGE2: 22.3621 ROUGEL: 29.1653 TOKENS_PER_SAMPLE: 272.8", - "metrics.result": 59505.3, - "metrics.result_per_accelerator": 7438.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1396 ROUGE2: 20.1904 ROUGEL: 30.0525 GEN_LEN: 4153772", + "metrics.result": 6821.11, + "metrics.result_per_accelerator": 852.63875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -18281,42 +19818,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": 2.4, + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", + "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.6325 ROUGE2: 23.3874 ROUGEL: 30.5099 TOKENS_PER_SAMPLE: 144.4 gsm8k_accuracy: 73.3 mbxp_accuracy: 60.18", - "metrics.result": 128795.0, - "metrics.result_per_accelerator": 16099.375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1295 ROUGE2: 20.1909 ROUGEL: 30.052 GEN_LEN: 4153653", + "metrics.result": 6823.91, + "metrics.result_per_accelerator": 852.98875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -18327,42 +19867,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": 2.4, + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", + "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4541 ROUGE2: 23.1973 ROUGEL: 30.2838 TOKENS_PER_SAMPLE: 146.3 gsm8k_accuracy: 72.94 mbxp_accuracy: 59.96", - "metrics.result": 129047.0, - "metrics.result_per_accelerator": 16130.875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1396 ROUGE2: 20.1904 ROUGEL: 30.0525 GEN_LEN: 4153772", + "metrics.result": 6821.11, + "metrics.result_per_accelerator": 852.63875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -18373,42 +19916,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.caches": "L1d: 6 MiB, L1i: 12 MiB, L2: 192 MiB, L3: 192 MiB", + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.frequency": 2.4, + "system.cpu.model": "Intel(R) Xeon(R) 6740E", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB 36ASF8G72PZ-3G2E1", + "system.name": "HPE ProLiant Compute DL380a Gen12 (8x L40S-PCIe-48GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", - "metrics.result": 98782.4, - "metrics.result_per_accelerator": 12347.8, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0194 ROUGE2: 20.0851 ROUGEL: 29.9991 GEN_LEN: 4155554", + "metrics.result": 7240.14, + "metrics.result_per_accelerator": 1810.035, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -18419,42 +19965,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.caches": "L1d: 9 MiB (192 instances), L1i: 6 MiB (192 instances), L2: 384 MiB (192 instances), L3: 420 MiB (4 instances)", + "system.cpu.core_count": 192, + "system.cpu.count_per_node": 4, + "system.cpu.frequency": 3.8, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 8192.0, + "system.memory.configuration": "1TB total, 32x 32GB", + "system.name": "HPE Compute Scale-up Server 3200 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7848 ROUGE2: 22.3844 ROUGEL: 29.1432 TOKENS_PER_SAMPLE: 273.8", - "metrics.result": 97629.8, - "metrics.result_per_accelerator": 12203.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0194 ROUGE2: 20.0851 ROUGEL: 29.9991 GEN_LEN: 4155554", + "metrics.result": 7179.86, + "metrics.result_per_accelerator": 1794.965, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -18465,41 +20014,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.5 LTS", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 180.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", + "system.cpu.caches": "L1d: 9 MiB (192 instances), L1i: 6 MiB (192 instances), L2: 384 MiB (192 instances), L3: 420 MiB (4 instances)", + "system.cpu.core_count": 192, + "system.cpu.count_per_node": 4, + "system.cpu.frequency": 3.8, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-5", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", - "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 8192.0, + "system.memory.configuration": "1TB total, 32x 32GB", + "system.name": "HPE Compute Scale-up Server 3200 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 18724.2, - "metrics.result_per_accelerator": 2340.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5453 ROUGE2: 22.1331 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.6", + "metrics.result": 33781.8, + "metrics.result_per_accelerator": 4222.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -18511,41 +20063,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6979P", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "SYS-522GA-NRT(8x H200-NVL-141GB)", + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 18571.1, - "metrics.result_per_accelerator": 2321.3875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5663 ROUGE2: 22.1467 ROUGEL: 28.7958 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 31281.1, + "metrics.result_per_accelerator": 3910.1375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -18557,40 +20112,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6979P", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "SYS-522GA-NRT(8x H200-NVL-141GB)", + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 18724.2, - "metrics.result_per_accelerator": 2340.525, + "metrics.result": 21014.6, + "metrics.result_per_accelerator": 2626.825, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -18603,40 +20161,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6979P", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "SYS-522GA-NRT(8x H200-NVL-141GB)", + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 18571.1, - "metrics.result_per_accelerator": 2321.3875, + "metrics.result": 21019.9, + "metrics.result_per_accelerator": 2627.4875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -18649,224 +20210,239 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6979P", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "SYS-522GA-NRT(8x H200-NVL-141GB)", + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.3988 ROUGE2: 22.039 ROUGEL: 28.6453 TOKENS_PER_SAMPLE: 295.3", - "metrics.result": 33069.1, - "metrics.result_per_accelerator": 4133.6375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21324.9, + "metrics.result_per_accelerator": 2665.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": null, - "software.framework": "vLLM 0.0.0.dev1+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.7.0", - "software.framework.rocm": "6.3.1", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.0.0", - "software.operating_system": "Ubuntu 22.04 LTS", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "RHEL 9.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 256.0, - "system.accelerator.memory_config": "HBM3E", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", - "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 6400MT/s @6000MT/s", - "system.name": "AS-8126GS-TNMR", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.46 ROUGE2: 22.0494 ROUGEL: 28.6787 TOKENS_PER_SAMPLE: 295.2", - "metrics.result": 28770.2, - "metrics.result_per_accelerator": 3596.275, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21025.5, + "metrics.result_per_accelerator": 2628.1875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": null, - "software.framework": "vLLM 0.0.0.dev1+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.7.0", - "software.framework.rocm": "6.3.1", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.0.0", - "software.operating_system": "Ubuntu 22.04 LTS", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "RHEL 9.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 256.0, - "system.accelerator.memory_config": "HBM3E", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", - "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 6400MT/s @6000MT/s", - "system.name": "AS-8126GS-TNMR", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.3988 ROUGE2: 22.039 ROUGEL: 28.6453 TOKENS_PER_SAMPLE: 295.3", - "metrics.result": 33069.1, - "metrics.result_per_accelerator": 4133.6375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34796.8, + "metrics.result_per_accelerator": 4349.6, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": null, - "software.framework": "vLLM 0.0.0.dev1+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.7.0", - "software.framework.rocm": "6.3.1", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.0.0", - "software.operating_system": "Ubuntu 22.04 LTS", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "RHEL 9.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 256.0, - "system.accelerator.memory_config": "HBM3E", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", - "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 6400MT/s @6000MT/s", - "system.name": "AS-8126GS-TNMR", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.46 ROUGE2: 22.0494 ROUGEL: 28.6787 TOKENS_PER_SAMPLE: 295.2", - "metrics.result": 28770.2, - "metrics.result_per_accelerator": 3596.275, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.497 ROUGE2: 22.0948 ROUGEL: 28.7299 TOKENS_PER_SAMPLE: 292.2", + "metrics.result": 19405.7, + "metrics.result_per_accelerator": 2425.7125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": null, - "software.framework": "vLLM 0.0.0.dev1+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.7.0", - "software.framework.rocm": "6.3.1", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.0.0", - "software.operating_system": "Ubuntu 22.04 LTS", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "RHEL 9.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 256.0, - "system.accelerator.memory_config": "HBM3E", - "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", - "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9575F", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 6400MT/s @6000MT/s", - "system.name": "AS-8126GS-TNMR", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6821 ROUGE2: 22.2379 ROUGEL: 28.8722 TOKENS_PER_SAMPLE: 289.4", - "metrics.result": 18084.9, - "metrics.result_per_accelerator": 2260.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34796.8, + "metrics.result_per_accelerator": 4349.6, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -18879,40 +20455,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6735 ROUGE2: 22.2241 ROUGEL: 28.8629 TOKENS_PER_SAMPLE: 290.3", - "metrics.result": 16848.6, - "metrics.result_per_accelerator": 2106.075, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.497 ROUGE2: 22.0948 ROUGEL: 28.7299 TOKENS_PER_SAMPLE: 292.2", + "metrics.result": 19405.7, + "metrics.result_per_accelerator": 2425.7125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -18925,41 +20504,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 14769.7, - "metrics.result_per_accelerator": 1846.2125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5453 ROUGE2: 22.133 ROUGEL: 28.7833 TOKENS_PER_SAMPLE: 291.6", + "metrics.result": 33964.0, + "metrics.result_per_accelerator": 4245.5, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -18971,41 +20553,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 14846.7, - "metrics.result_per_accelerator": 1855.8375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5588 ROUGE2: 22.1412 ROUGEL: 28.7922 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 31283.3, + "metrics.result_per_accelerator": 3910.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -19017,41 +20602,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "RHEL 9.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.caches": "L1d: 3 MiB (64 instances), L1i: 2 MiB (64 instances), L2: 128 MiB (64 instances), L3: 120 MiB (2 instances)", "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": 2.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.memory.configuration": "32x HPE Cray Supercomputing XD 64GB (1x64GB) Dual Rank x4 DDR5-5600 Registered Standard Memory", + "system.name": "HPE Cray XD670 with Cray ClusterStor (8x H200-SXM-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 14769.7, - "metrics.result_per_accelerator": 1846.2125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5484 ROUGE2: 22.1341 ROUGEL: 28.7828 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 9362.85, + "metrics.result_per_accelerator": 4681.425, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -19063,41 +20651,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 14846.7, - "metrics.result_per_accelerator": 1855.8375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5503 ROUGE2: 22.1345 ROUGEL: 28.7855 TOKENS_PER_SAMPLE: 292.5", + "metrics.result": 8674.58, + "metrics.result_per_accelerator": 4337.29, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -19109,41 +20700,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6743 ROUGE2: 22.2269 ROUGEL: 28.8649 TOKENS_PER_SAMPLE: 289.3", - "metrics.result": 15553.3, - "metrics.result_per_accelerator": 1944.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 5666.16, + "metrics.result_per_accelerator": 2833.08, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -19155,41 +20749,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6131 ROUGE2: 22.198 ROUGEL: 28.8436 TOKENS_PER_SAMPLE: 290.5", - "metrics.result": 11041.5, - "metrics.result_per_accelerator": 1380.1875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 5039.82, + "metrics.result_per_accelerator": 2519.91, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -19201,41 +20798,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6743 ROUGE2: 22.2269 ROUGEL: 28.8649 TOKENS_PER_SAMPLE: 289.3", - "metrics.result": 15553.3, - "metrics.result_per_accelerator": 1944.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 5666.16, + "metrics.result_per_accelerator": 2833.08, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -19247,41 +20847,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Fujitsu", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 8, + "submission.division": "closed", + "submission.organization": "HPE", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6131 ROUGE2: 22.198 ROUGEL: 28.8436 TOKENS_PER_SAMPLE: 290.5", - "metrics.result": 11041.5, - "metrics.result_per_accelerator": 1380.1875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 5039.82, + "metrics.result_per_accelerator": 2519.91, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -19293,36 +20896,39 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.451 ROUGE2: 23.2629 ROUGEL: 30.3646 TOKENS_PER_SAMPLE: 144.9 gsm8k_accuracy: 73.42 mbxp_accuracy: 59.9", - "metrics.result": 37713.6, - "metrics.result_per_accelerator": 4714.2, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4581 ROUGE2: 23.2422 ROUGEL: 30.2949 TOKENS_PER_SAMPLE: 145.4 gsm8k_accuracy: 73.22 mbxp_accuracy: 59.88", + "metrics.result": 16703.4, + "metrics.result_per_accelerator": 8351.7, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "mixtral-8x7b", @@ -19339,36 +20945,39 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4195 ROUGE2: 23.275 ROUGEL: 30.3242 TOKENS_PER_SAMPLE: 146.4 gsm8k_accuracy: 73.04 mbxp_accuracy: 59.94", - "metrics.result": 37934.5, - "metrics.result_per_accelerator": 4741.8125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.385 ROUGE2: 23.2684 ROUGEL: 30.2469 TOKENS_PER_SAMPLE: 146.2 gsm8k_accuracy: 72.94 mbxp_accuracy: 59.8", + "metrics.result": 15570.8, + "metrics.result_per_accelerator": 7785.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "mixtral-8x7b", @@ -19385,36 +20994,39 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6821 ROUGE2: 22.2379 ROUGEL: 28.8722 TOKENS_PER_SAMPLE: 289.4", - "metrics.result": 18084.9, - "metrics.result_per_accelerator": 2260.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5484 ROUGE2: 22.1341 ROUGEL: 28.7828 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 9362.85, + "metrics.result_per_accelerator": 4681.425, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", @@ -19431,36 +21043,39 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6735 ROUGE2: 22.2241 ROUGEL: 28.8629 TOKENS_PER_SAMPLE: 290.3", - "metrics.result": 16848.6, - "metrics.result_per_accelerator": 2106.075, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5503 ROUGE2: 22.1345 ROUGEL: 28.7855 TOKENS_PER_SAMPLE: 292.5", + "metrics.result": 8674.58, + "metrics.result_per_accelerator": 4337.29, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", @@ -19477,36 +21092,39 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 24.04.1 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "HPE", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 144.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", - "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (2x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5269 ROUGE2: 22.1161 ROUGEL: 28.7612 TOKENS_PER_SAMPLE: 290.6", - "metrics.result": 17654.1, - "metrics.result_per_accelerator": 4413.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34264.0, + "metrics.result_per_accelerator": 4283.0, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", @@ -19526,33 +21144,36 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 4, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9655 96-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB2", - "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB RAM", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5272 ROUGE2: 22.1154 ROUGEL: 28.7622 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 14652.8, - "metrics.result_per_accelerator": 3663.2, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5513 ROUGE2: 22.1359 ROUGEL: 28.7885 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 33027.3, + "metrics.result_per_accelerator": 4128.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", @@ -19572,33 +21193,36 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 4, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9655 96-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB2", - "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB RAM", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 10825.5, - "metrics.result_per_accelerator": 2706.375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20999.6, + "metrics.result_per_accelerator": 2624.95, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", @@ -19618,33 +21242,36 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 4, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9655 96-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB2", - "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB RAM", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 10503.7, - "metrics.result_per_accelerator": 2625.925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21359.6, + "metrics.result_per_accelerator": 2669.95, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", @@ -19664,33 +21291,36 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 4, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9655 96-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB2", - "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB RAM", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 10825.5, - "metrics.result_per_accelerator": 2706.375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20999.6, + "metrics.result_per_accelerator": 2624.95, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", @@ -19710,33 +21340,36 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 4, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9655 96-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB2", - "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB RAM", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 10503.7, - "metrics.result_per_accelerator": 2625.925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21359.6, + "metrics.result_per_accelerator": 2669.95, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", @@ -19756,33 +21389,36 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 4, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9655 96-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB2", - "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB RAM", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4435 ROUGE2: 23.2094 ROUGEL: 30.2857 TOKENS_PER_SAMPLE: 145.3 gsm8k_accuracy: 73.3 mbxp_accuracy: 59.86", - "metrics.result": 31503.7, - "metrics.result_per_accelerator": 7875.925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4064 ROUGE2: 23.1865 ROUGEL: 30.2546 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.14", + "metrics.result": 60923.3, + "metrics.result_per_accelerator": 7615.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "mixtral-8x7b", @@ -19802,33 +21438,36 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 4, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9655 96-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB2", - "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB RAM", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.416 ROUGE2: 23.2389 ROUGEL: 30.246 TOKENS_PER_SAMPLE: 146.8 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", - "metrics.result": 29301.5, - "metrics.result_per_accelerator": 7325.375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4575 ROUGE2: 23.2383 ROUGEL: 30.3743 TOKENS_PER_SAMPLE: 145.8 gsm8k_accuracy: 73.26 mbxp_accuracy: 59.76", + "metrics.result": 57054.4, + "metrics.result_per_accelerator": 7131.8, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "mixtral-8x7b", @@ -19848,39 +21487,42 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Sustainable_Metal_Cloud", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 4, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9655 96-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB2", - "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB RAM", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5269 ROUGE2: 22.1161 ROUGEL: 28.7612 TOKENS_PER_SAMPLE: 290.6", - "metrics.result": 17654.1, - "metrics.result_per_accelerator": 4413.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", + "metrics.result": 97124.0, + "metrics.result_per_accelerator": 12140.5, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -19891,42 +21533,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", + "software.operating_system": "Debian 12", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 4, + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9655 96-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB2", - "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5272 ROUGE2: 22.1154 ROUGEL: 28.7622 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 14652.8, - "metrics.result_per_accelerator": 3663.2, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7884 ROUGE2: 22.3874 ROUGEL: 29.148 TOKENS_PER_SAMPLE: 273.8", + "metrics.result": 92338.7, + "metrics.result_per_accelerator": 11542.3375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -19937,42 +21582,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", + "software.operating_system": "Debian 12", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 4, + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9655 96-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB2", - "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 7705.97, - "metrics.result_per_accelerator": 1926.4925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 22.092192082812325 exact_match: 90.10742885506286 TOKENS_PER_SAMPLE: 654.1", + "metrics.result": 1498.56, + "metrics.result_per_accelerator": 187.32, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -19983,42 +21631,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", + "software.operating_system": "Debian 12", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Lenovo", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "submission.organization": "Google", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 8.1 MiB, L1i: 10.8 MiB, L2: 344 MiB, L3: 672 MiB", - "system.cpu.core_count": 86, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6787P", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94BDJHA464N", - "system.name": "ThinkSystem SR650a V4 (4x H100-NVL-94GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 6773.87, - "metrics.result_per_accelerator": 1693.4675, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 22.048188322528272 exact_match: 90.08918266048973 TOKENS_PER_SAMPLE: 656.1", + "metrics.result": 847.484, + "metrics.result_per_accelerator": 105.9355, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -20029,42 +21680,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", + "software.operating_system": "Debian 12", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 8.1 MiB, L1i: 10.8 MiB, L2: 344 MiB, L3: 672 MiB", - "system.cpu.core_count": 86, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6787P", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94BDJHA464N", - "system.name": "ThinkSystem SR650a V4 (4x H100-NVL-94GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 7705.97, - "metrics.result_per_accelerator": 1926.4925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4539 ROUGE2: 23.217 ROUGEL: 30.2997 TOKENS_PER_SAMPLE: 144.9 gsm8k_accuracy: 73.16 mbxp_accuracy: 60.5", + "metrics.result": 121489.0, + "metrics.result_per_accelerator": 15186.125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -20075,42 +21729,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", + "software.operating_system": "Debian 12", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 8.1 MiB, L1i: 10.8 MiB, L2: 344 MiB, L3: 672 MiB", - "system.cpu.core_count": 86, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6787P", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94BDJHA464N", - "system.name": "ThinkSystem SR650a V4 (4x H100-NVL-94GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 6773.87, - "metrics.result_per_accelerator": 1693.4675, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4486 ROUGE2: 23.1917 ROUGEL: 30.2808 TOKENS_PER_SAMPLE: 146.4 gsm8k_accuracy: 73.12 mbxp_accuracy: 60.4", + "metrics.result": 122411.0, + "metrics.result_per_accelerator": 15301.375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -20121,42 +21778,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", + "software.operating_system": "Debian 12", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 8.1 MiB, L1i: 10.8 MiB, L2: 344 MiB, L3: 672 MiB", - "system.cpu.core_count": 86, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6787P", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94BDJHA464N", - "system.name": "ThinkSystem SR650a V4 (4x H100-NVL-94GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3468 ROUGE2: 23.202 ROUGEL: 30.3039 TOKENS_PER_SAMPLE: 145.4 gsm8k_accuracy: 73.34 mbxp_accuracy: 60.02", - "metrics.result": 19500.9, - "metrics.result_per_accelerator": 4875.225, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", + "metrics.result": 97089.9, + "metrics.result_per_accelerator": 12136.2375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -20167,42 +21827,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", + "software.operating_system": "Debian 12", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 8.1 MiB, L1i: 10.8 MiB, L2: 344 MiB, L3: 672 MiB", - "system.cpu.core_count": 86, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6787P", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94BDJHA464N", - "system.name": "ThinkSystem SR650a V4 (4x H100-NVL-94GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4582 ROUGE2: 23.1907 ROUGEL: 30.36 TOKENS_PER_SAMPLE: 146.9 gsm8k_accuracy: 72.98 mbxp_accuracy: 59.74", - "metrics.result": 19334.9, - "metrics.result_per_accelerator": 4833.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7895 ROUGE2: 22.3823 ROUGEL: 29.1462 TOKENS_PER_SAMPLE: 273.8", + "metrics.result": 92203.7, + "metrics.result_per_accelerator": 11525.4625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -20213,36 +21876,39 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", - "submission.availability": "available", + "software.operating_system": "Debian 12", + "submission.availability": "preview", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 8.1 MiB, L1i: 10.8 MiB, L2: 344 MiB, L3: 672 MiB", - "system.cpu.core_count": 86, + "system.cpu.caches": null, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) 6787P", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94BDJHA464N", - "system.name": "ThinkSystem SR650a V4 (4x H100-NVL-94GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7628 TOKENS_PER_SAMPLE: 290.7", - "metrics.result": 35400.2, - "metrics.result_per_accelerator": 4425.025, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34144.7, + "metrics.result_per_accelerator": 4268.0875, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", @@ -20262,7 +21928,7 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -20271,24 +21937,27 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": 2.1, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", + "system.cpu.vcpu_count": "224", "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5214 ROUGE2: 22.1163 ROUGEL: 28.7576 TOKENS_PER_SAMPLE: 291.6", - "metrics.result": 32934.7, - "metrics.result_per_accelerator": 4116.8375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5593 ROUGE2: 22.1417 ROUGEL: 28.7894 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 31390.2, + "metrics.result_per_accelerator": 3923.775, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", @@ -20308,7 +21977,7 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -20317,24 +21986,27 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": 2.1, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", + "system.cpu.vcpu_count": "224", "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 21626.0, - "metrics.result_per_accelerator": 2703.25, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21040.9, + "metrics.result_per_accelerator": 2630.1125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", @@ -20354,7 +22026,7 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -20363,24 +22035,27 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": 2.1, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", + "system.cpu.vcpu_count": "224", "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 21553.4, - "metrics.result_per_accelerator": 2694.175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20501.8, + "metrics.result_per_accelerator": 2562.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", @@ -20400,7 +22075,7 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -20409,24 +22084,27 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": 2.1, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", + "system.cpu.vcpu_count": "224", "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 21626.0, - "metrics.result_per_accelerator": 2703.25, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21040.9, + "metrics.result_per_accelerator": 2630.1125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", @@ -20446,7 +22124,7 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -20455,24 +22133,27 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": 2.1, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", + "system.cpu.vcpu_count": "224", "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 21553.4, - "metrics.result_per_accelerator": 2694.175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20501.8, + "metrics.result_per_accelerator": 2562.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", @@ -20492,7 +22173,7 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -20501,29 +22182,32 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": 2.1, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", + "system.cpu.vcpu_count": "224", "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7629 TOKENS_PER_SAMPLE: 290.7", - "metrics.result": 35498.5, - "metrics.result_per_accelerator": 4437.3125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.688728239185064 exact_match: 89.9708041032429 TOKENS_PER_SAMPLE: 645.8", + "metrics.result": 552.592, + "metrics.result_per_accelerator": 69.074, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -20538,7 +22222,7 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -20547,29 +22231,32 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": 2.1, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", + "system.cpu.vcpu_count": "224", "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.576 ROUGE2: 22.1459 ROUGEL: 28.7741 TOKENS_PER_SAMPLE: 292.6", - "metrics.result": 19448.1, - "metrics.result_per_accelerator": 2431.0125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.677865835546807 exact_match: 90.07546988749174 TOKENS_PER_SAMPLE: 648.2", + "metrics.result": 276.43, + "metrics.result_per_accelerator": 34.55375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -20584,7 +22271,7 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -20593,29 +22280,32 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": 2.1, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", + "system.cpu.vcpu_count": "224", "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7629 TOKENS_PER_SAMPLE: 290.7", - "metrics.result": 35498.5, - "metrics.result_per_accelerator": 4437.3125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4064 ROUGE2: 23.1865 ROUGEL: 30.2546 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.14", + "metrics.result": 61614.5, + "metrics.result_per_accelerator": 7701.8125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -20630,7 +22320,7 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -20639,29 +22329,32 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": 2.1, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", + "system.cpu.vcpu_count": "224", "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.576 ROUGE2: 22.1459 ROUGEL: 28.7741 TOKENS_PER_SAMPLE: 292.6", - "metrics.result": 19448.1, - "metrics.result_per_accelerator": 2431.0125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.3268 ROUGE2: 23.1698 ROUGEL: 30.2987 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.48 mbxp_accuracy: 59.78", + "metrics.result": 59100.1, + "metrics.result_per_accelerator": 7387.5125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -20676,7 +22369,7 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -20685,29 +22378,32 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": 2.1, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", + "system.cpu.vcpu_count": "224", "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4086 ROUGE2: 23.1851 ROUGEL: 30.262 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.12", - "metrics.result": 63514.7, - "metrics.result_per_accelerator": 7939.3375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34144.7, + "metrics.result_per_accelerator": 4268.0875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -20722,7 +22418,7 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -20731,29 +22427,32 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": 2.1, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", + "system.cpu.vcpu_count": "224", "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4078 ROUGE2: 23.1492 ROUGEL: 30.338 TOKENS_PER_SAMPLE: 144.7 gsm8k_accuracy: 73.02 mbxp_accuracy: 60.0", - "metrics.result": 61074.2, - "metrics.result_per_accelerator": 7634.275, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5593 ROUGE2: 22.1417 ROUGEL: 28.7894 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 31390.2, + "metrics.result_per_accelerator": 3923.775, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -20768,7 +22467,7 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Google", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -20777,213 +22476,228 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", - "system.cpu.vcpu_count": null, + "system.cpu.frequency": 2.1, + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8581C CPU", + "system.cpu.vcpu_count": "224", "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "a3-ultragpu-8g (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7628 TOKENS_PER_SAMPLE: 290.7", - "metrics.result": 35400.2, - "metrics.result_per_accelerator": 4425.025, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4701 ROUGE2: 22.0744 ROUGEL: 28.6808 TOKENS_PER_SAMPLE: 294.8", + "metrics.result": 103182.0, + "metrics.result_per_accelerator": 3224.4375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "ROCm 6.3.0", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.rocm": "6.3.0", "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.5 LTS (jammy)", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "MangoBoost", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.total_count": 32, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "AMD EPYC 9534", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", - "system.number_of_nodes": 1, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 2252.8, + "system.memory.configuration": "24x 96GB Samsung M321RYGA0BB0-CQKZJ", + "system.name": "MangoBoost Mi300X (32x MI300X-192GB-HBM3, LLMBoost)", + "system.number_of_nodes": 4, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5214 ROUGE2: 22.1163 ROUGEL: 28.7576 TOKENS_PER_SAMPLE: 291.6", - "metrics.result": 32934.7, - "metrics.result_per_accelerator": 4116.8375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4557 ROUGE2: 22.0578 ROUGEL: 28.672 TOKENS_PER_SAMPLE: 295.1", + "metrics.result": 93039.9, + "metrics.result_per_accelerator": 2907.496875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "ROCm 6.3.0", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.rocm": "6.3.0", "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.5 LTS (jammy)", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "MangoBoost", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.total_count": 32, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "AMD EPYC 9534", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", - "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", - "system.number_of_nodes": 1, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 2252.8, + "system.memory.configuration": "24x 96GB Samsung M321RYGA0BB0-CQKZJ", + "system.name": "MangoBoost Mi300X (32x MI300X-192GB-HBM3, LLMBoost)", + "system.number_of_nodes": 4, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7628 TOKENS_PER_SAMPLE: 290.7", - "metrics.result": 35453.0, - "metrics.result_per_accelerator": 4431.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4701 ROUGE2: 22.0744 ROUGEL: 28.6808 TOKENS_PER_SAMPLE: 294.8", + "metrics.result": 103182.0, + "metrics.result_per_accelerator": 3224.4375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "ROCm 6.3.0", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.rocm": "6.3.0", "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.5 LTS (jammy)", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "MangoBoost", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.total_count": 32, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "AMD EPYC 9534", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", - "system.number_of_nodes": 1, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 2252.8, + "system.memory.configuration": "24x 96GB Samsung M321RYGA0BB0-CQKZJ", + "system.name": "MangoBoost Mi300X (32x MI300X-192GB-HBM3, LLMBoost)", + "system.number_of_nodes": 4, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5225 ROUGE2: 22.1121 ROUGEL: 28.7551 TOKENS_PER_SAMPLE: 291.6", - "metrics.result": 32976.4, - "metrics.result_per_accelerator": 4122.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4557 ROUGE2: 22.0578 ROUGEL: 28.672 TOKENS_PER_SAMPLE: 295.1", + "metrics.result": 93039.9, + "metrics.result_per_accelerator": 2907.496875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "ROCm 6.3.0", "software.framework.ai": null, - "software.framework.cuda": "12.8", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.rocm": "6.3.0", "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Ubuntu 22.04.5 LTS (jammy)", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "MangoBoost", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.memory_capacity": 192.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI300X 192GB HBM3", + "system.accelerator.total_count": 32, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "AMD EPYC 9534", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", - "system.number_of_nodes": 1, + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": null, + "system.memory.capacity": 2252.8, + "system.memory.configuration": "24x 96GB Samsung M321RYGA0BB0-CQKZJ", + "system.name": "MangoBoost Mi300X (32x MI300X-192GB-HBM3, LLMBoost)", + "system.number_of_nodes": 4, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 21513.3, - "metrics.result_per_accelerator": 2689.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5042 ROUGE2: 22.0887 ROUGEL: 28.749 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 3144.18, + "metrics.result_per_accelerator": 3144.18, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -20995,41 +22709,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 21549.1, - "metrics.result_per_accelerator": 2693.6375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5019 ROUGE2: 22.0906 ROUGEL: 28.7528 TOKENS_PER_SAMPLE: 292.5", + "metrics.result": 3203.87, + "metrics.result_per_accelerator": 3203.87, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21041,40 +22758,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 21513.3, - "metrics.result_per_accelerator": 2689.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 2808.21, + "metrics.result_per_accelerator": 2808.21, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -21087,40 +22807,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", - "metrics.result": 21549.1, - "metrics.result_per_accelerator": 2693.6375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 2824.55, + "metrics.result_per_accelerator": 2824.55, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -21133,41 +22856,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7628 TOKENS_PER_SAMPLE: 290.7", - "metrics.result": 35383.8, - "metrics.result_per_accelerator": 4422.975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 2808.21, + "metrics.result_per_accelerator": 2808.21, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21179,41 +22905,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5762 ROUGE2: 22.1484 ROUGEL: 28.7736 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 19492.5, - "metrics.result_per_accelerator": 2436.5625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1186 ROUGE2: 20.1722 ROUGEL: 30.0185 GEN_LEN: 4156655", + "metrics.result": 2824.55, + "metrics.result_per_accelerator": 2824.55, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21225,41 +22954,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7628 TOKENS_PER_SAMPLE: 290.7", - "metrics.result": 35383.8, - "metrics.result_per_accelerator": 4422.975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4156 ROUGE2: 23.1543 ROUGEL: 30.2816 TOKENS_PER_SAMPLE: 145.2 gsm8k_accuracy: 72.94 mbxp_accuracy: 59.74", + "metrics.result": 7792.91, + "metrics.result_per_accelerator": 7792.91, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21271,41 +23003,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5762 ROUGE2: 22.1484 ROUGEL: 28.7736 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 19492.5, - "metrics.result_per_accelerator": 2436.5625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.3856 ROUGE2: 23.246 ROUGEL: 30.2962 TOKENS_PER_SAMPLE: 146.2 gsm8k_accuracy: 73.38 mbxp_accuracy: 59.82", + "metrics.result": 7771.75, + "metrics.result_per_accelerator": 7771.75, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21317,41 +23052,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4074 ROUGE2: 23.2168 ROUGEL: 30.2703 TOKENS_PER_SAMPLE: 145.3 gsm8k_accuracy: 73.62 mbxp_accuracy: 60.24", - "metrics.result": 63446.7, - "metrics.result_per_accelerator": 7930.8375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5042 ROUGE2: 22.0887 ROUGEL: 28.749 TOKENS_PER_SAMPLE: 291.3", + "metrics.result": 3144.18, + "metrics.result_per_accelerator": 3144.18, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21363,41 +23101,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3802 ROUGE2: 23.1803 ROUGEL: 30.3072 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.0 mbxp_accuracy: 60.2", - "metrics.result": 61317.2, - "metrics.result_per_accelerator": 7664.65, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5019 ROUGE2: 22.0906 ROUGEL: 28.7528 TOKENS_PER_SAMPLE: 292.5", + "metrics.result": 3203.87, + "metrics.result_per_accelerator": 3203.87, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21409,40 +23150,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.3", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "QuantaGrid S74G-2U (1x GH200-96GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7628 TOKENS_PER_SAMPLE: 290.7", - "metrics.result": 35453.0, - "metrics.result_per_accelerator": 4431.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5767 ROUGE2: 22.1512 ROUGEL: 28.773 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 31083.6, + "metrics.result_per_accelerator": 3885.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -21455,40 +23199,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5225 ROUGE2: 22.1121 ROUGEL: 28.7551 TOKENS_PER_SAMPLE: 291.6", - "metrics.result": 32976.4, - "metrics.result_per_accelerator": 4122.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5665 ROUGE2: 22.1466 ROUGEL: 28.7638 TOKENS_PER_SAMPLE: 292.9", + "metrics.result": 29916.0, + "metrics.result_per_accelerator": 3739.5, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -21501,45 +23248,48 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB M321R8GA0PB0", - "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3218 ROUGE2: 23.243 ROUGEL: 30.2997 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.08 mbxp_accuracy: 59.9", - "metrics.result": 51514.5, - "metrics.result_per_accelerator": 6439.3125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20464.9, + "metrics.result_per_accelerator": 2558.1125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, @@ -21547,45 +23297,48 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1, VMware ESXi 8.0.3 2441450", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "Virtualized NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "Intel Xeon Platinum 8468", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "TBD", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB,VMware ESXi 8.0.3) ", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.5359 ROUGE2: 23.2614 ROUGEL: 30.4398 TOKENS_PER_SAMPLE: 146.4 gsm8k_accuracy: 73.12 mbxp_accuracy: 59.92", - "metrics.result": 50896.0, - "metrics.result_per_accelerator": 6362.0, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 19316.9, + "metrics.result_per_accelerator": 2414.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, @@ -21593,41 +23346,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04.1, VMware ESXi 8.0.3 2441450", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "Virtualized NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.1, - "system.cpu.model": "Intel Xeon Platinum 8468", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "TBD", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB,VMware ESXi 8.0.3) ", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5457 ROUGE2: 22.133 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34763.2, - "metrics.result_per_accelerator": 4345.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20464.9, + "metrics.result_per_accelerator": 2558.1125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21639,41 +23395,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5602 ROUGE2: 22.142 ROUGEL: 28.794 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 33023.5, - "metrics.result_per_accelerator": 4127.9375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 19316.9, + "metrics.result_per_accelerator": 2414.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21685,41 +23444,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21406.8, - "metrics.result_per_accelerator": 2675.85, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.562456891920377 exact_match: 90.11248841826605 TOKENS_PER_SAMPLE: 638.5", + "metrics.result": 436.218, + "metrics.result_per_accelerator": 54.52725, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21731,41 +23493,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21455.9, - "metrics.result_per_accelerator": 2681.9875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.565117800148553 exact_match: 90.08828921244209 TOKENS_PER_SAMPLE: 624.9", + "metrics.result": 261.98, + "metrics.result_per_accelerator": 32.7475, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21777,41 +23542,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21406.8, - "metrics.result_per_accelerator": 2675.85, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4914 ROUGE2: 23.3009 ROUGEL: 30.3779 TOKENS_PER_SAMPLE: 145.2 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", + "metrics.result": 51048.7, + "metrics.result_per_accelerator": 6381.0875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21823,41 +23591,44 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21455.9, - "metrics.result_per_accelerator": 2681.9875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4321 ROUGE2: 23.1829 ROUGEL: 30.3742 TOKENS_PER_SAMPLE: 146.5 gsm8k_accuracy: 73.86 mbxp_accuracy: 59.88", + "metrics.result": 50924.7, + "metrics.result_per_accelerator": 6365.5875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -21869,40 +23640,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34538.4, - "metrics.result_per_accelerator": 4317.3, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5767 ROUGE2: 22.1512 ROUGEL: 28.773 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 31083.6, + "metrics.result_per_accelerator": 3885.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -21915,40 +23689,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5133 ROUGE2: 22.1131 ROUGEL: 28.7404 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 19415.9, - "metrics.result_per_accelerator": 2426.9875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5665 ROUGE2: 22.1466 ROUGEL: 28.7638 TOKENS_PER_SAMPLE: 292.9", + "metrics.result": 29916.0, + "metrics.result_per_accelerator": 3739.5, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -21961,43 +23738,46 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D74H-7U (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, - { - "metrics.accuracy": "ROUGEL: 21.618544328060732 exact_match: 89.99261085373924 TOKENS_PER_SAMPLE: 646.3", - "metrics.result": 559.205, - "metrics.result_per_accelerator": 69.900625, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5773 ROUGE2: 22.1546 ROUGEL: 28.7713 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 7013.17, + "metrics.result_per_accelerator": 1753.2925, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -22005,45 +23785,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.8.0.43", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.67381013091784 exact_match: 89.96315023163469 TOKENS_PER_SAMPLE: 650.6", - "metrics.result": 291.486, - "metrics.result_per_accelerator": 36.43575, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6283 ROUGE2: 22.2205 ROUGEL: 28.8297 TOKENS_PER_SAMPLE: 290.7", + "metrics.result": 4889.92, + "metrics.result_per_accelerator": 1222.48, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama3.1-405b", - "model.name": "llama3_1-405b", - "model.number_of_parameters": 405.0, + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -22051,45 +23834,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.8.0.43", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34538.4, - "metrics.result_per_accelerator": 4317.3, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1385 ROUGE2: 20.1564 ROUGEL: 30.0269 GEN_LEN: 4155440", + "metrics.result": 6319.07, + "metrics.result_per_accelerator": 1579.7675, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -22097,45 +23883,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.8.0.43", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5133 ROUGE2: 22.1131 ROUGEL: 28.7404 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 19415.9, - "metrics.result_per_accelerator": 2426.9875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1385 ROUGE2: 20.1564 ROUGEL: 30.0269 GEN_LEN: 4155440", + "metrics.result": 6251.27, + "metrics.result_per_accelerator": 1562.8175, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -22143,45 +23932,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.8.0.43", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.37 ROUGE2: 23.2496 ROUGEL: 30.2632 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.3 mbxp_accuracy: 60.1", - "metrics.result": 62042.2, - "metrics.result_per_accelerator": 7755.275, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1385 ROUGE2: 20.1564 ROUGEL: 30.0269 GEN_LEN: 4155440", + "metrics.result": 6319.07, + "metrics.result_per_accelerator": 1579.7675, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -22189,45 +23981,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.8.0.43", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.4712 ROUGE2: 23.1931 ROUGEL: 30.3668 TOKENS_PER_SAMPLE: 146.4 gsm8k_accuracy: 73.2 mbxp_accuracy: 60.06", - "metrics.result": 58972.6, - "metrics.result_per_accelerator": 7371.575, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1385 ROUGE2: 20.1564 ROUGEL: 30.0269 GEN_LEN: 4155440", + "metrics.result": 6251.27, + "metrics.result_per_accelerator": 1562.8175, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -22235,45 +24030,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.8.0.43", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5457 ROUGE2: 22.133 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34763.2, - "metrics.result_per_accelerator": 4345.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.3734 ROUGE2: 23.2194 ROUGEL: 30.3241 TOKENS_PER_SAMPLE: 145.4 gsm8k_accuracy: 73.22 mbxp_accuracy: 59.86", + "metrics.result": 15843.3, + "metrics.result_per_accelerator": 3960.825, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -22281,45 +24079,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.8.0.43", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5602 ROUGE2: 22.142 ROUGEL: 28.794 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 33023.5, - "metrics.result_per_accelerator": 4127.9375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.3176 ROUGE2: 23.1416 ROUGEL: 30.2829 TOKENS_PER_SAMPLE: 145.4 gsm8k_accuracy: 73.52 mbxp_accuracy: 60.02", + "metrics.result": 13864.5, + "metrics.result_per_accelerator": 3466.125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -22327,45 +24128,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.8.0.43", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 4096.0, - "system.memory.configuration": "32x 128GB DDR5", - "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34822.6, - "metrics.result_per_accelerator": 4352.825, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5773 ROUGE2: 22.1546 ROUGEL: 28.7713 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 7013.17, + "metrics.result_per_accelerator": 1753.2925, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -22373,45 +24177,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.8.0.43", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5569 ROUGE2: 22.145 ROUGEL: 28.7929 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 33051.9, - "metrics.result_per_accelerator": 4131.4875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6283 ROUGE2: 22.2205 ROUGEL: 28.8297 TOKENS_PER_SAMPLE: 290.7", + "metrics.result": 4889.92, + "metrics.result_per_accelerator": 1222.48, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework": "TensorRT 10.8.0.43, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", "software.framework.cudnn": null, @@ -22419,44 +24226,47 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.8", + "software.framework.tensorrt": "10.8.0.43", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 24.04", + "software.operating_system": "Rocky Linux 9.2", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Quanta_Cloud_Technology", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM2e", + "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 52, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": " DDR4-4800 64GB x 32", + "system.name": "QuantaGrid D54U-3U (4x H100-PCIe-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21117.0, - "metrics.result_per_accelerator": 2639.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7466 ROUGE2: 22.3524 ROUGEL: 29.1548 TOKENS_PER_SAMPLE: 271.7", + "metrics.result": 12269.3, + "metrics.result_per_accelerator": 12269.3, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -22470,39 +24280,42 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (1x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21536.4, - "metrics.result_per_accelerator": 2692.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7469 ROUGE2: 22.3525 ROUGEL: 29.1541 TOKENS_PER_SAMPLE: 272.7", + "metrics.result": 12078.4, + "metrics.result_per_accelerator": 12078.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -22516,39 +24329,42 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (1x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21117.0, - "metrics.result_per_accelerator": 2639.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7466 ROUGE2: 22.3524 ROUGEL: 29.1548 TOKENS_PER_SAMPLE: 271.7", + "metrics.result": 12269.3, + "metrics.result_per_accelerator": 12269.3, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -22562,39 +24378,42 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (1x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21536.4, - "metrics.result_per_accelerator": 2692.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7469 ROUGE2: 22.3525 ROUGEL: 29.1541 TOKENS_PER_SAMPLE: 272.7", + "metrics.result": 12078.4, + "metrics.result_per_accelerator": 12078.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -22608,37 +24427,40 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (1x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34545.5, - "metrics.result_per_accelerator": 4318.1875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5775 ROUGE2: 22.1518 ROUGEL: 28.7746 TOKENS_PER_SAMPLE: 291.8", + "metrics.result": 31306.8, + "metrics.result_per_accelerator": 3913.35, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -22654,37 +24476,40 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5036 ROUGE2: 22.0997 ROUGEL: 28.733 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 19428.4, - "metrics.result_per_accelerator": 2428.55, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5683 ROUGE2: 22.1486 ROUGEL: 28.7643 TOKENS_PER_SAMPLE: 292.9", + "metrics.result": 31106.3, + "metrics.result_per_accelerator": 3888.2875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -22700,37 +24525,40 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34545.5, - "metrics.result_per_accelerator": 4318.1875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5775 ROUGE2: 22.1518 ROUGEL: 28.7746 TOKENS_PER_SAMPLE: 291.8", + "metrics.result": 31306.8, + "metrics.result_per_accelerator": 3913.35, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -22746,37 +24574,40 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5036 ROUGE2: 22.0997 ROUGEL: 28.733 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 19428.4, - "metrics.result_per_accelerator": 2428.55, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5683 ROUGE2: 22.1486 ROUGEL: 28.7643 TOKENS_PER_SAMPLE: 292.9", + "metrics.result": 31106.3, + "metrics.result_per_accelerator": 3888.2875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -22792,37 +24623,40 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 34822.6, - "metrics.result_per_accelerator": 4352.825, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5449 ROUGE2: 22.134 ROUGEL: 28.7831 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 4686.14, + "metrics.result_per_accelerator": 4686.14, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -22838,37 +24672,40 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 144, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5569 ROUGE2: 22.145 ROUGEL: 28.7929 TOKENS_PER_SAMPLE: 292.3", - "metrics.result": 33051.9, - "metrics.result_per_accelerator": 4131.4875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5529 ROUGE2: 22.1393 ROUGEL: 28.7909 TOKENS_PER_SAMPLE: 292.5", + "metrics.result": 4350.26, + "metrics.result_per_accelerator": 4350.26, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -22884,38 +24721,41 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 144, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB", - "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 31149.9, - "metrics.result_per_accelerator": 3893.7375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 2824.83, + "metrics.result_per_accelerator": 2824.83, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -22930,38 +24770,41 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 144, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5583 ROUGE2: 22.1382 ROUGEL: 28.7896 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 28848.4, - "metrics.result_per_accelerator": 3606.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 2526.75, + "metrics.result_per_accelerator": 2526.75, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -22976,37 +24819,40 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 144, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 16900.1, - "metrics.result_per_accelerator": 2112.5125, + "metrics.result": 2824.83, + "metrics.result_per_accelerator": 2824.83, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -23022,37 +24868,40 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 144, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 17975.8, - "metrics.result_per_accelerator": 2246.975, + "metrics.result": 2526.75, + "metrics.result_per_accelerator": 2526.75, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -23068,38 +24917,41 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 144, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 16900.1, - "metrics.result_per_accelerator": 2112.5125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.485 ROUGE2: 23.27 ROUGEL: 30.3119 TOKENS_PER_SAMPLE: 145.1 gsm8k_accuracy: 73.48 mbxp_accuracy: 59.94", + "metrics.result": 8345.73, + "metrics.result_per_accelerator": 8345.73, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -23114,38 +24966,41 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 144, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 17975.8, - "metrics.result_per_accelerator": 2246.975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4444 ROUGE2: 23.2749 ROUGEL: 30.3007 TOKENS_PER_SAMPLE: 146.3 gsm8k_accuracy: 73.24 mbxp_accuracy: 59.88", + "metrics.result": 8042.27, + "metrics.result_per_accelerator": 8042.27, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, @@ -23160,37 +25015,40 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 144, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 31334.8, - "metrics.result_per_accelerator": 3916.85, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5449 ROUGE2: 22.134 ROUGEL: 28.7831 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 4686.14, + "metrics.result_per_accelerator": 4686.14, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -23206,37 +25064,40 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 144, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5859 ROUGE2: 22.1562 ROUGEL: 28.8145 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 16134.4, - "metrics.result_per_accelerator": 2016.8, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5529 ROUGE2: 22.1393 ROUGEL: 28.7909 TOKENS_PER_SAMPLE: 292.5", + "metrics.result": 4350.26, + "metrics.result_per_accelerator": 4350.26, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -23252,39 +25113,42 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 144.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, - "system.cpu.count_per_node": 2, + "system.cpu.core_count": 144, + "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", + "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 31334.8, - "metrics.result_per_accelerator": 3916.85, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7454 ROUGE2: 22.3503 ROUGEL: 29.1532 TOKENS_PER_SAMPLE: 271.7", + "metrics.result": 98858.0, + "metrics.result_per_accelerator": 12357.25, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23298,39 +25162,42 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5859 ROUGE2: 22.1562 ROUGEL: 28.8145 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 16134.4, - "metrics.result_per_accelerator": 2016.8, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7462 ROUGE2: 22.3556 ROUGEL: 29.1559 TOKENS_PER_SAMPLE: 272.6", + "metrics.result": 98443.3, + "metrics.result_per_accelerator": 12305.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23344,39 +25211,42 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", - "metrics.result": 31149.9, - "metrics.result_per_accelerator": 3893.7375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7454 ROUGE2: 22.3503 ROUGEL: 29.1532 TOKENS_PER_SAMPLE: 271.7", + "metrics.result": 98858.0, + "metrics.result_per_accelerator": 12357.25, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23390,39 +25260,42 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5583 ROUGE2: 22.1382 ROUGEL: 28.7896 TOKENS_PER_SAMPLE: 292.4", - "metrics.result": 28848.4, - "metrics.result_per_accelerator": 3606.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.827 ROUGE2: 22.4067 ROUGEL: 29.208 TOKENS_PER_SAMPLE: 272.5", + "metrics.result": 59409.3, + "metrics.result_per_accelerator": 7426.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23436,39 +25309,42 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-NVL-141GB", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 192, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5751 ROUGE2: 22.1504 ROUGEL: 28.7726 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 31216.8, - "metrics.result_per_accelerator": 3902.1, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.985172700345778 exact_match: 90.12750165453342 TOKENS_PER_SAMPLE: 655.9", + "metrics.result": 1526.29, + "metrics.result_per_accelerator": 190.78625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23479,42 +25355,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5724 ROUGE2: 22.1518 ROUGEL: 28.7699 TOKENS_PER_SAMPLE: 292.9", - "metrics.result": 30903.6, - "metrics.result_per_accelerator": 3862.95, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.975590204593704 exact_match: 90.07971211118463 TOKENS_PER_SAMPLE: 654.0", + "metrics.result": 845.82, + "metrics.result_per_accelerator": 105.7275, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23525,42 +25404,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20663.7, - "metrics.result_per_accelerator": 2582.9625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7454 ROUGE2: 22.3503 ROUGEL: 29.1532 TOKENS_PER_SAMPLE: 271.7", + "metrics.result": 98858.0, + "metrics.result_per_accelerator": 12357.25, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23571,42 +25453,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21134.4, - "metrics.result_per_accelerator": 2641.8, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.827 ROUGE2: 22.4067 ROUGEL: 29.208 TOKENS_PER_SAMPLE: 272.5", + "metrics.result": 59409.3, + "metrics.result_per_accelerator": 7426.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23617,42 +25502,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 20663.7, - "metrics.result_per_accelerator": 2582.9625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.5969 ROUGE2: 23.2676 ROUGEL: 30.3668 TOKENS_PER_SAMPLE: 143.9 gsm8k_accuracy: 73.08 mbxp_accuracy: 60.3", + "metrics.result": 128148.0, + "metrics.result_per_accelerator": 16018.5, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23663,42 +25551,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", - "metrics.result": 21134.4, - "metrics.result_per_accelerator": 2641.8, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4131 ROUGE2: 23.183 ROUGEL: 30.2308 TOKENS_PER_SAMPLE: 145.5 gsm8k_accuracy: 73.24 mbxp_accuracy: 59.7", + "metrics.result": 126845.0, + "metrics.result_per_accelerator": 15855.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp8", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23709,42 +25600,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5767 ROUGE2: 22.1512 ROUGEL: 28.773 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 30989.5, - "metrics.result_per_accelerator": 3873.6875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7454 ROUGE2: 22.3503 ROUGEL: 29.1532 TOKENS_PER_SAMPLE: 271.7", + "metrics.result": 98858.0, + "metrics.result_per_accelerator": 12357.25, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23755,42 +25649,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5968 ROUGE2: 22.1606 ROUGEL: 28.7638 TOKENS_PER_SAMPLE: 293.1", - "metrics.result": 12979.4, - "metrics.result_per_accelerator": 1622.425, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7462 ROUGE2: 22.3556 ROUGEL: 29.1559 TOKENS_PER_SAMPLE: 272.6", + "metrics.result": 98443.3, + "metrics.result_per_accelerator": 12305.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99", - "model.name": "llama2-70b-interactive-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23801,42 +25698,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA179N", + "system.name": "NVIDIA DGX B200 (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.643467134956637 exact_match: 90.07990403706154 TOKENS_PER_SAMPLE: 645.6", - "metrics.result": 397.31, - "metrics.result_per_accelerator": 49.66375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.93672502698354 exact_match: 90.12851091992057 TOKENS_PER_SAMPLE: 656.4", + "metrics.result": 13886.4, + "metrics.result_per_accelerator": 192.86666666666667, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama3.1-405b", "model.name": "llama3_1-405b", "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23847,42 +25747,45 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 186.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GB200", + "system.accelerator.total_count": 72, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", - "system.number_of_nodes": 1, + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "for each 4-GPU node: 1.3TB LPDDR5x", + "system.name": " (72x GB200-186GB_aarch64, TensorRT, Triton)", + "system.number_of_nodes": 18, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGEL: 21.578037465211967 exact_match: 90.02991065519522 TOKENS_PER_SAMPLE: 646.3", - "metrics.result": 277.125, - "metrics.result_per_accelerator": 34.640625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.959911702352596 exact_match: 90.1118795499669 TOKENS_PER_SAMPLE: 656.8", + "metrics.result": 8850.1, + "metrics.result_per_accelerator": 122.91805555555555, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama3.1-405b", "model.name": "llama3_1-405b", "model.number_of_parameters": 405.0, - "model.weight_data_types": "fp8", + "model.weight_data_types": "fp4", "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, "software.framework.cuda": "12.8", @@ -23893,40 +25796,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 186.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA GB200", + "system.accelerator.total_count": 72, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", - "system.number_of_nodes": 1, + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "for each 4-GPU node: 1.3TB LPDDR5x", + "system.name": " (72x GB200-186GB_aarch64, TensorRT, Triton)", + "system.number_of_nodes": 18, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5767 ROUGE2: 22.1512 ROUGEL: 28.773 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 30989.5, - "metrics.result_per_accelerator": 3873.6875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5388 ROUGE2: 22.127 ROUGEL: 28.7729 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 34300.2, + "metrics.result_per_accelerator": 4287.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -23939,40 +25845,43 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5968 ROUGE2: 22.1606 ROUGEL: 28.7638 TOKENS_PER_SAMPLE: 293.1", - "metrics.result": 12979.4, - "metrics.result_per_accelerator": 1622.425, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5568 ROUGE2: 22.1417 ROUGEL: 28.7946 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 33024.7, + "metrics.result_per_accelerator": 4128.0875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-interactive-99.9", - "model.name": "llama2-70b-interactive-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -23985,36 +25894,39 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5751 ROUGE2: 22.1504 ROUGEL: 28.7726 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 31216.8, - "metrics.result_per_accelerator": 3902.1, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5388 ROUGE2: 22.127 ROUGEL: 28.7729 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 34300.2, + "metrics.result_per_accelerator": 4287.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", @@ -24031,36 +25943,39 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5724 ROUGE2: 22.1518 ROUGEL: 28.7699 TOKENS_PER_SAMPLE: 292.9", - "metrics.result": 30903.6, - "metrics.result_per_accelerator": 3862.95, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5568 ROUGE2: 22.1417 ROUGEL: 28.7946 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 33024.7, + "metrics.result_per_accelerator": 4128.0875, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", @@ -24077,36 +25992,39 @@ "software.framework.sima": null, "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5645 ROUGE2: 22.1587 ROUGEL: 28.7857 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 3481.53, - "metrics.result_per_accelerator": 435.19125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5459 ROUGE2: 22.1333 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34988.2, + "metrics.result_per_accelerator": 4373.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", @@ -24126,33 +26044,36 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 384, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "PCIe-5", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB", - "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5496 ROUGE2: 22.1516 ROUGEL: 28.7732 TOKENS_PER_SAMPLE: 294.1", - "metrics.result": 3201.13, - "metrics.result_per_accelerator": 400.14125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1489 ROUGEL: 28.797 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 33071.5, + "metrics.result_per_accelerator": 4133.9375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", @@ -24172,33 +26093,36 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 384, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "PCIe-5", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB", - "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0584 ROUGE2: 20.1404 ROUGEL: 29.9742 GEN_LEN: 4125446", - "metrics.result": 6946.68, - "metrics.result_per_accelerator": 868.335, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21182.5, + "metrics.result_per_accelerator": 2647.8125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", @@ -24218,33 +26142,36 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 384, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "PCIe-5", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB", - "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.055 ROUGE2: 20.1229 ROUGEL: 29.9616 GEN_LEN: 4127336", - "metrics.result": 6213.34, - "metrics.result_per_accelerator": 776.6675, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21544.6, + "metrics.result_per_accelerator": 2693.075, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", @@ -24264,33 +26191,36 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 384, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "PCIe-5", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB", - "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0584 ROUGE2: 20.1404 ROUGEL: 29.9742 GEN_LEN: 4125446", - "metrics.result": 6946.68, - "metrics.result_per_accelerator": 868.335, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21182.5, + "metrics.result_per_accelerator": 2647.8125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", @@ -24310,33 +26240,36 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 384, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "PCIe-5", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB", - "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.055 ROUGE2: 20.1229 ROUGEL: 29.9616 GEN_LEN: 4127336", - "metrics.result": 6213.34, - "metrics.result_per_accelerator": 776.6675, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21544.6, + "metrics.result_per_accelerator": 2693.075, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", @@ -24356,37 +26289,40 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 384, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "PCIe-5", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB", - "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5645 ROUGE2: 22.1587 ROUGEL: 28.7857 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 3481.53, - "metrics.result_per_accelerator": 435.19125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34989.2, + "metrics.result_per_accelerator": 4373.65, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -24402,37 +26338,40 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 384, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "PCIe-5", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB", - "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5496 ROUGE2: 22.1516 ROUGEL: 28.7732 TOKENS_PER_SAMPLE: 294.1", - "metrics.result": 3201.13, - "metrics.result_per_accelerator": 400.14125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5031 ROUGE2: 22.1008 ROUGEL: 28.7296 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 19424.2, + "metrics.result_per_accelerator": 2428.025, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", "software.framework": "TensorRT 10.8, CUDA 12.8", @@ -24448,3319 +26387,3290 @@ "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 384, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9965 192-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "PCIe-5", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB", - "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5321 ROUGE2: 22.1191 ROUGEL: 28.7225 TOKENS_PER_SAMPLE: 293.7", - "metrics.result": 3062.72, - "metrics.result_per_accelerator": 3062.72, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.63486183033167 exact_match: 90.05081072137656 TOKENS_PER_SAMPLE: 648.6", + "metrics.result": 560.651, + "metrics.result_per_accelerator": 70.081375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.2", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "AMD", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", - "system.accelerator.total_count": 1, - "system.accelerator.vendor": "AMD", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2xAMD EPYC 9374F", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", - "system.name": "Supermicro AS-8125GS-TNMR2", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5105 ROUGE2: 22.0812 ROUGEL: 28.727 TOKENS_PER_SAMPLE: 292.7", - "metrics.result": 2520.27, - "metrics.result_per_accelerator": 2520.27, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.6585777555619 exact_match: 90.07448378557245 TOKENS_PER_SAMPLE: 647.2", + "metrics.result": 291.374, + "metrics.result_per_accelerator": 36.42175, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.2", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "AMD", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", - "system.accelerator.total_count": 1, - "system.accelerator.vendor": "AMD", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2xAMD EPYC 9374F", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", - "system.name": "Supermicro AS-8125GS-TNMR2", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5321 ROUGE2: 22.1191 ROUGEL: 28.7225 TOKENS_PER_SAMPLE: 293.7", - "metrics.result": 3062.72, - "metrics.result_per_accelerator": 3062.72, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34989.2, + "metrics.result_per_accelerator": 4373.65, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.2", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "AMD", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", - "system.accelerator.total_count": 1, - "system.accelerator.vendor": "AMD", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2xAMD EPYC 9374F", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", - "system.name": "Supermicro AS-8125GS-TNMR2", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5105 ROUGE2: 22.0812 ROUGEL: 28.727 TOKENS_PER_SAMPLE: 292.7", - "metrics.result": 2520.27, - "metrics.result_per_accelerator": 2520.27, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5031 ROUGE2: 22.1008 ROUGEL: 28.7296 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 19424.2, + "metrics.result_per_accelerator": 2428.025, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.2", - "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", + "software.framework.pytorch": null, + "software.framework.rocm": null, + "software.framework.sima": null, + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "AMD", + "submission.organization": "NVIDIA", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", - "system.accelerator.total_count": 1, - "system.accelerator.vendor": "AMD", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2xAMD EPYC 9374F", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", - "system.name": "Supermicro AS-8125GS-TNMR2", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5633 ROUGE2: 22.1252 ROUGEL: 28.7479 TOKENS_PER_SAMPLE: 292.7", - "metrics.result": 23514.8, - "metrics.result_per_accelerator": 2939.35, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4064 ROUGE2: 23.1865 ROUGEL: 30.2546 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.14", + "metrics.result": 62629.6, + "metrics.result_per_accelerator": 7828.7, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.2", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "AMD", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2xAMD EPYC 9374F", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", - "system.name": "Supermicro AS-8125GS-TNMR2", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5522 ROUGE2: 22.1184 ROUGEL: 28.7592 TOKENS_PER_SAMPLE: 293.5", - "metrics.result": 21028.2, - "metrics.result_per_accelerator": 2628.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4092 ROUGE2: 23.1612 ROUGEL: 30.3199 TOKENS_PER_SAMPLE: 144.4 gsm8k_accuracy: 73.4 mbxp_accuracy: 59.66", + "metrics.result": 61802.0, + "metrics.result_per_accelerator": 7725.25, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.2", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "AMD", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2xAMD EPYC 9374F", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", - "system.name": "Supermicro AS-8125GS-TNMR2", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5633 ROUGE2: 22.1252 ROUGEL: 28.7479 TOKENS_PER_SAMPLE: 292.7", - "metrics.result": 23514.8, - "metrics.result_per_accelerator": 2939.35, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5459 ROUGE2: 22.1333 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34988.2, + "metrics.result_per_accelerator": 4373.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.2", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "AMD", + "submission.organization": "NVIDIA", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2xAMD EPYC 9374F", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", - "system.name": "Supermicro AS-8125GS-TNMR2", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5522 ROUGE2: 22.1184 ROUGEL: 28.7592 TOKENS_PER_SAMPLE: 293.5", - "metrics.result": 21028.2, - "metrics.result_per_accelerator": 2628.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1489 ROUGEL: 28.797 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 33071.5, + "metrics.result_per_accelerator": 4133.9375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.2", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.4 LTS (Jammy Jellyfish)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "AMD", + "submission.organization": "NVIDIA", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2xAMD EPYC 9374F", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "XGMI", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB Micron MTC40F2046S1RC48BA1 MHCC", - "system.name": "Supermicro AS-8125GS-TNMR2", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5633 ROUGE2: 22.1252 ROUGEL: 28.7479 TOKENS_PER_SAMPLE: 292.7", - "metrics.result": 24109.8, - "metrics.result_per_accelerator": 3013.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.389 ROUGE2: 22.0197 ROUGEL: 28.6092 TOKENS_PER_SAMPLE: 297.4", + "metrics.result": 33928.1, + "metrics.result_per_accelerator": 4241.0125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", + "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.2", + "software.framework.pytorch": "2.7.0", + "software.framework.rocm": "6.3.1", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.6 LTS (Jammy Jellyfish)", - "submission.availability": "preview", + "software.framework.vllm": "0.6.5", + "software.operating_system": "Ubuntu 22.04.5 LTS", + "submission.availability": "available", "submission.division": "closed", "submission.organization": "AMD", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_config": "HBM3E", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", "system.accelerator.total_count": 8, "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": null, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2xAMD EPYC TURIN", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB Samsung M321R8GA0PB1-CCPXC", - "system.name": "Supermicro AS-8125GS-TNMR2", + "system.memory.capacity": 2304.0, + "system.memory.configuration": "24x 96GiB Samsung M321RYGA0PB0-CWMXJ", + "system.name": "QuantaGrid D74A-7U", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5555 ROUGE2: 22.0993 ROUGEL: 28.7204 TOKENS_PER_SAMPLE: 292.8", - "metrics.result": 22020.9, - "metrics.result_per_accelerator": 2752.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4508 ROUGE2: 22.058 ROUGEL: 28.6575 TOKENS_PER_SAMPLE: 295.8", + "metrics.result": 30724.5, + "metrics.result_per_accelerator": 3840.5625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", + "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.2", + "software.framework.pytorch": "2.7.0", + "software.framework.rocm": "6.3.1", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.6 LTS (Jammy Jellyfish)", - "submission.availability": "preview", + "software.framework.vllm": "0.6.5", + "software.operating_system": "Ubuntu 22.04.5 LTS", + "submission.availability": "available", "submission.division": "closed", "submission.organization": "AMD", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_config": "HBM3E", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", "system.accelerator.total_count": 8, "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": null, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2xAMD EPYC TURIN", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB Samsung M321R8GA0PB1-CCPXC", - "system.name": "Supermicro AS-8125GS-TNMR2", + "system.memory.capacity": 2304.0, + "system.memory.configuration": "24x 96GiB Samsung M321RYGA0PB0-CWMXJ", + "system.name": "QuantaGrid D74A-7U", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5633 ROUGE2: 22.1252 ROUGEL: 28.7479 TOKENS_PER_SAMPLE: 292.7", - "metrics.result": 24109.8, - "metrics.result_per_accelerator": 3013.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.389 ROUGE2: 22.0197 ROUGEL: 28.6092 TOKENS_PER_SAMPLE: 297.4", + "metrics.result": 33928.1, + "metrics.result_per_accelerator": 4241.0125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", + "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.2", + "software.framework.pytorch": "2.7.0", + "software.framework.rocm": "6.3.1", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.6 LTS (Jammy Jellyfish)", - "submission.availability": "preview", + "software.framework.vllm": "0.6.5", + "software.operating_system": "Ubuntu 22.04.5 LTS", + "submission.availability": "available", "submission.division": "closed", "submission.organization": "AMD", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_config": "HBM3E", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", "system.accelerator.total_count": 8, "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": null, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2xAMD EPYC TURIN", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB Samsung M321R8GA0PB1-CCPXC", - "system.name": "Supermicro AS-8125GS-TNMR2", + "system.memory.capacity": 2304.0, + "system.memory.configuration": "24x 96GiB Samsung M321RYGA0PB0-CWMXJ", + "system.name": "QuantaGrid D74A-7U", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5555 ROUGE2: 22.0993 ROUGEL: 28.7204 TOKENS_PER_SAMPLE: 292.8", - "metrics.result": 22020.9, - "metrics.result_per_accelerator": 2752.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4508 ROUGE2: 22.058 ROUGEL: 28.6575 TOKENS_PER_SAMPLE: 295.8", + "metrics.result": 30724.5, + "metrics.result_per_accelerator": 3840.5625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.2", + "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", "software.framework.ai": null, "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.2", + "software.framework.pytorch": "2.7.0", + "software.framework.rocm": "6.3.1", "software.framework.sima": null, "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.6 LTS (Jammy Jellyfish)", - "submission.availability": "preview", + "software.framework.vllm": "0.6.5", + "software.operating_system": "Ubuntu 22.04.5 LTS", + "submission.availability": "available", "submission.division": "closed", "submission.organization": "AMD", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD Instinct MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_config": "HBM3E", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", "system.accelerator.total_count": 8, "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": null, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "2xAMD EPYC TURIN", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB Samsung M321R8GA0PB1-CCPXC", - "system.name": "Supermicro AS-8125GS-TNMR2", - "system.number_of_nodes": 1, - "system.type": "datacenter" - }, - { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 6985.51, - "metrics.result_per_accelerator": 1746.3775, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", - "software.framework.ai": null, - "software.framework.cuda": "12.4", - "software.framework.cudnn": null, - "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", - "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "ASUSTeK", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654 96-Core Processor", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 768.0, - "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", - "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", - "system.number_of_nodes": 1, - "system.type": "datacenter" - }, - { - "metrics.accuracy": "ROUGE1: 43.1008 ROUGE2: 20.1119 ROUGEL: 29.9747 GEN_LEN: 4114209", - "metrics.result": 6070.26, - "metrics.result_per_accelerator": 1517.565, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", - "software.framework.ai": null, - "software.framework.cuda": "12.4", - "software.framework.cudnn": null, - "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", - "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "ASUSTeK", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654 96-Core Processor", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 768.0, - "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", - "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", + "system.memory.capacity": 2304.0, + "system.memory.configuration": "24x 96GiB Samsung M321RYGA0PB0-CWMXJ", + "system.name": "QuantaGrid D74A-7U", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 6985.51, - "metrics.result_per_accelerator": 1746.3775, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34556.6, + "metrics.result_per_accelerator": 4319.575, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5 LT", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Oracle", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654 96-Core Processor", + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 2.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 768.0, - "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", - "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1008 ROUGE2: 20.1119 ROUGEL: 29.9747 GEN_LEN: 4114209", - "metrics.result": 6070.26, - "metrics.result_per_accelerator": 1517.565, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5513 ROUGE2: 22.1381 ROUGEL: 28.7884 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 33001.0, + "metrics.result_per_accelerator": 4125.125, "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.architecture": "LLM", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5 LT", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Oracle", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654 96-Core Processor", + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 2.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 768.0, - "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", - "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5664 ROUGE2: 22.1601 ROUGEL: 28.7722 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 6673.95, - "metrics.result_per_accelerator": 1668.4875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21104.1, + "metrics.result_per_accelerator": 2638.0125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5 LT", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Oracle", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654 96-Core Processor", + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 2.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 768.0, - "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", - "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5664 ROUGE2: 22.16 ROUGEL: 28.7714 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 5788.29, - "metrics.result_per_accelerator": 1447.0725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21527.4, + "metrics.result_per_accelerator": 2690.925, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5 LT", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Oracle", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654 96-Core Processor", + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 2.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 768.0, - "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", - "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5664 ROUGE2: 22.1601 ROUGEL: 28.7722 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 6673.95, - "metrics.result_per_accelerator": 1668.4875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21104.1, + "metrics.result_per_accelerator": 2638.0125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5 LT", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Oracle", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654 96-Core Processor", + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 2.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 768.0, - "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", - "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5664 ROUGE2: 22.16 ROUGEL: 28.7714 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 5788.29, - "metrics.result_per_accelerator": 1447.0725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21527.4, + "metrics.result_per_accelerator": 2690.925, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5 LT", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Oracle", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654 96-Core Processor", + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 2.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 768.0, - "system.memory.configuration": "12x64GB MTC40F2046S1RC48BA1", - "system.name": "ASUSTeK ESC4000A-E12 (4xH100-NVL-94GB)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1261 ROUGE2: 20.1655 ROUGEL: 30.025 GEN_LEN: 4101272", - "metrics.result": 13078.5, - "metrics.result_per_accelerator": 1634.8125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4064 ROUGE2: 23.1865 ROUGEL: 30.2546 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.12", + "metrics.result": 60708.6, + "metrics.result_per_accelerator": 7588.575, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5 LT", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Oracle", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", + "system.cpu.frequency": 2.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB ", - "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.126 ROUGE2: 20.1652 ROUGEL: 30.026 GEN_LEN: 4101320", - "metrics.result": 7986.46, - "metrics.result_per_accelerator": 998.3075, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.3758 ROUGE2: 23.2396 ROUGEL: 30.3046 TOKENS_PER_SAMPLE: 146.3 gsm8k_accuracy: 73.72 mbxp_accuracy: 60.0", + "metrics.result": 59149.8, + "metrics.result_per_accelerator": 7393.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5 LT", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Oracle", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", + "system.cpu.frequency": 2.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB ", - "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1261 ROUGE2: 20.1655 ROUGEL: 30.025 GEN_LEN: 4101272", - "metrics.result": 13078.5, - "metrics.result_per_accelerator": 1634.8125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34556.6, + "metrics.result_per_accelerator": 4319.575, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5 LT", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Oracle", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", + "system.cpu.frequency": 2.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB ", - "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.126 ROUGE2: 20.1652 ROUGEL: 30.026 GEN_LEN: 4101320", - "metrics.result": 7986.46, - "metrics.result_per_accelerator": 998.3075, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5513 ROUGE2: 22.1381 ROUGEL: 28.7884 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 33001.0, + "metrics.result_per_accelerator": 4125.125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5 LT", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "ASUSTeK", + "submission.organization": "Oracle", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", + "system.cpu.frequency": 2.0, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB ", - "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.566 ROUGE2: 22.16 ROUGEL: 28.7719 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 9281.5, - "metrics.result_per_accelerator": 1160.1875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.3988 ROUGE2: 22.039 ROUGEL: 28.6453 TOKENS_PER_SAMPLE: 295.3", + "metrics.result": 33366.5, + "metrics.result_per_accelerator": 4170.8125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": null, + "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.7.0", + "software.framework.rocm": "6.3.1", "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.6.5", + "software.operating_system": "22.04.5 LTS (Jammy Jellyfish)", "submission.availability": "available", "submission.division": "closed", "submission.organization": "ASUSTeK", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", + "system.cpu.model": "2x AMD EPYC 9475F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB ", - "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2252.8, + "system.memory.configuration": "24x 96GB Samsung M321RYGA0PB2-CCPPC", + "system.name": "ASUSTeK ESC A8A-E12U (8x MI325x-256GB )", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5654 ROUGE2: 22.1595 ROUGEL: 28.7714 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 8094.27, - "metrics.result_per_accelerator": 1011.78375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.3971 ROUGE2: 22.0537 ROUGEL: 28.6495 TOKENS_PER_SAMPLE: 295.7", + "metrics.result": 30086.5, + "metrics.result_per_accelerator": 3760.8125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": null, + "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.7.0", + "software.framework.rocm": "6.3.1", "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.6.5", + "software.operating_system": "22.04.5 LTS (Jammy Jellyfish)", "submission.availability": "available", "submission.division": "closed", "submission.organization": "ASUSTeK", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", + "system.cpu.model": "2x AMD EPYC 9475F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB ", - "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2252.8, + "system.memory.configuration": "24x 96GB Samsung M321RYGA0PB2-CCPPC", + "system.name": "ASUSTeK ESC A8A-E12U (8x MI325x-256GB )", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.566 ROUGE2: 22.16 ROUGEL: 28.7719 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 9281.5, - "metrics.result_per_accelerator": 1160.1875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.3988 ROUGE2: 22.039 ROUGEL: 28.6453 TOKENS_PER_SAMPLE: 295.3", + "metrics.result": 33366.5, + "metrics.result_per_accelerator": 4170.8125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": null, + "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.7.0", + "software.framework.rocm": "6.3.1", "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.6.5", + "software.operating_system": "22.04.5 LTS (Jammy Jellyfish)", "submission.availability": "available", "submission.division": "closed", "submission.organization": "ASUSTeK", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", + "system.cpu.model": "2x AMD EPYC 9475F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB ", - "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2252.8, + "system.memory.configuration": "24x 96GB Samsung M321RYGA0PB2-CCPPC", + "system.name": "ASUSTeK ESC A8A-E12U (8x MI325x-256GB )", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5654 ROUGE2: 22.1595 ROUGEL: 28.7714 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 8094.27, - "metrics.result_per_accelerator": 1011.78375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.3971 ROUGE2: 22.0537 ROUGEL: 28.6495 TOKENS_PER_SAMPLE: 295.7", + "metrics.result": 30086.5, + "metrics.result_per_accelerator": 3760.8125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": null, + "software.framework": "vLLM 0.6.5.dev964+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.7.0", + "software.framework.rocm": "6.3.1", "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.6.5", + "software.operating_system": "22.04.5 LTS (Jammy Jellyfish)", "submission.availability": "available", "submission.division": "closed", "submission.organization": "ASUSTeK", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", + "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9374F 32-Core Processor", + "system.cpu.model": "2x AMD EPYC 9475F", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "16x 64GB ", - "system.name": "ESC8000A-E12 (8x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2252.8, + "system.memory.configuration": "24x 96GB Samsung M321RYGA0PB2-CCPPC", + "system.name": "ASUSTeK ESC A8A-E12U (8x MI325x-256GB )", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 19877.6, - "metrics.result_per_accelerator": 2484.7, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5429 ROUGE2: 22.1318 ROUGEL: 28.7758 TOKENS_PER_SAMPLE: 291.4", + "metrics.result": 27765.3, + "metrics.result_per_accelerator": 3470.6625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", "submission.organization": "ASUSTeK", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", + "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", - "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB ", + "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0994 ROUGE2: 20.1115 ROUGEL: 29.9734 GEN_LEN: 4114014", - "metrics.result": 19226.9, - "metrics.result_per_accelerator": 2403.3625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5628 ROUGE2: 22.1477 ROUGEL: 28.7948 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 27417.3, + "metrics.result_per_accelerator": 3427.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", "submission.organization": "ASUSTeK", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", + "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", - "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB ", + "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 19877.6, - "metrics.result_per_accelerator": 2484.7, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", + "metrics.result": 18113.4, + "metrics.result_per_accelerator": 2264.175, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", "submission.organization": "ASUSTeK", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", + "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", - "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB ", + "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0994 ROUGE2: 20.1115 ROUGEL: 29.9734 GEN_LEN: 4114014", - "metrics.result": 19226.9, - "metrics.result_per_accelerator": 2403.3625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", + "metrics.result": 17953.9, + "metrics.result_per_accelerator": 2244.2375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", "submission.organization": "ASUSTeK", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", + "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", - "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB ", + "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5663 ROUGE2: 22.1602 ROUGEL: 28.7722 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 24323.6, - "metrics.result_per_accelerator": 3040.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", + "metrics.result": 18113.4, + "metrics.result_per_accelerator": 2264.175, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", "submission.organization": "ASUSTeK", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", + "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", - "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB ", + "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5663 ROUGE2: 22.1597 ROUGEL: 28.7718 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 20605.3, - "metrics.result_per_accelerator": 2575.6625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", + "metrics.result": 17953.9, + "metrics.result_per_accelerator": 2244.2375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", "submission.organization": "ASUSTeK", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", + "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", - "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB ", + "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5663 ROUGE2: 22.1602 ROUGEL: 28.7722 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 24323.6, - "metrics.result_per_accelerator": 3040.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5429 ROUGE2: 22.1318 ROUGEL: 28.7758 TOKENS_PER_SAMPLE: 291.4", + "metrics.result": 27765.3, + "metrics.result_per_accelerator": 3470.6625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", "submission.organization": "ASUSTeK", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", + "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", - "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB ", + "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5663 ROUGE2: 22.1597 ROUGEL: 28.7718 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 20605.3, - "metrics.result_per_accelerator": 2575.6625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5628 ROUGE2: 22.1477 ROUGEL: 28.7948 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 27417.3, + "metrics.result_per_accelerator": 3427.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", "submission.organization": "ASUSTeK", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 64, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8592+", + "system.cpu.model": "AMD EPYC 9475F 48-Core Processo", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "16x 96GB M321RYGA0PB2-CCPPC", - "system.name": "ESC-N8-E11 (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB ", + "system.name": "ASUSTeK ESC8000A-E13P (8x H200-NVL-141GB NVLink, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0843 ROUGE2: 20.0853 ROUGEL: 29.9346 GEN_LEN: 4114897", - "metrics.result": 1747.67, - "metrics.result_per_accelerator": 873.835, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5468 ROUGE2: 22.134 ROUGEL: 28.7841 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34807.1, + "metrics.result_per_accelerator": 4350.8875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Cisco", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 2, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel Xeon Gold 6448H", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", - "system.name": "Cisco UCS C240 M7 (2x L40S, TensorRT)", - "system.number_of_nodes": 1, - "system.type": "datacenter" - }, - { - "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.0866 ROUGEL: 29.9354 GEN_LEN: 4114953", - "metrics.result": 1734.03, - "metrics.result_per_accelerator": 867.015, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", - "software.framework.ai": null, - "software.framework.cuda": "12.4", - "software.framework.cudnn": null, - "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", - "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel Xeon Gold 6448H", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", - "system.name": "Cisco UCS C240 M7 (2x L40S, TensorRT)", - "system.number_of_nodes": 1, - "system.type": "datacenter" - }, - { - "metrics.accuracy": "ROUGE1: 43.0362 ROUGE2: 20.0871 ROUGEL: 29.9403 GEN_LEN: 4123417", - "metrics.result": 3268.94, - "metrics.result_per_accelerator": 1634.47, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", - "software.framework.ai": null, - "software.framework.cuda": "12.4", - "software.framework.cudnn": null, - "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", - "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "Cisco", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 2, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9684X 96-Core Processor", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", - "system.name": "Cisco UCS C245 M8 (2x H100-PCIe-80GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0354 ROUGE2: 20.0871 ROUGEL: 29.9398 GEN_LEN: 4123289", - "metrics.result": 3258.83, - "metrics.result_per_accelerator": 1629.415, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5621 ROUGE2: 22.1419 ROUGEL: 28.7948 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 33037.2, + "metrics.result_per_accelerator": 4129.65, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Cisco", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9684X 96-Core Processor", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", - "system.name": "Cisco UCS C245 M8 (2x H100-PCIe-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0605 ROUGE2: 20.1151 ROUGEL: 29.9703 GEN_LEN: 4112035", - "metrics.result": 1729.6, - "metrics.result_per_accelerator": 864.8, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", + "metrics.result": 21171.3, + "metrics.result_per_accelerator": 2646.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Cisco", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9684X 96-Core Processor", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", - "system.name": "Cisco UCS C245 M8 (2x L40S, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0621 ROUGE2: 20.1162 ROUGEL: 29.9713 GEN_LEN: 4112159", - "metrics.result": 1725.54, - "metrics.result_per_accelerator": 862.77, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", + "metrics.result": 21533.9, + "metrics.result_per_accelerator": 2691.7375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Cisco", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9684X 96-Core Processor", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", - "system.name": "Cisco UCS C245 M8 (2x L40S, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0605 ROUGE2: 20.1151 ROUGEL: 29.9703 GEN_LEN: 4112035", - "metrics.result": 1746.46, - "metrics.result_per_accelerator": 873.23, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", + "metrics.result": 21171.3, + "metrics.result_per_accelerator": 2646.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Cisco", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8562Y+", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", - "system.name": "Cisco UCS X210c M7 (2x L40S, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0605 ROUGE2: 20.1151 ROUGEL: 29.9703 GEN_LEN: 4112037", - "metrics.result": 1725.98, - "metrics.result_per_accelerator": 862.99, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1301 ROUGE2: 20.133 ROUGEL: 30.0141 GEN_LEN: 4157005", + "metrics.result": 21533.9, + "metrics.result_per_accelerator": 2691.7375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Cisco", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8562Y+", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", - "system.name": "Cisco UCS X210c M7 (2x L40S, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0605 ROUGE2: 20.1151 ROUGEL: 29.9703 GEN_LEN: 4112035", - "metrics.result": 1746.46, - "metrics.result_per_accelerator": 873.23, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34809.3, + "metrics.result_per_accelerator": 4351.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Cisco", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8562Y+", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", - "system.name": "Cisco UCS X210c M7 (2x L40S, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0605 ROUGE2: 20.1151 ROUGEL: 29.9703 GEN_LEN: 4112037", - "metrics.result": 1725.98, - "metrics.result_per_accelerator": 862.99, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5018 ROUGE2: 22.0993 ROUGEL: 28.7329 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 19418.5, + "metrics.result_per_accelerator": 2427.3125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Cisco", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8562Y+", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB HMCG94MEBRA109N", - "system.name": "Cisco UCS X210c M7 (2x L40S, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 42.9363 ROUGE2: 20.1275 ROUGEL: 29.9464 GEN_LEN: 4048420", - "metrics.result": 64.0078, - "metrics.result_per_accelerator": 64.0078, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "Jetpack 6.0, TensorRT 10.1, CUDA 12.2", - "software.framework.ai": null, - "software.framework.cuda": "12.2", - "software.framework.cudnn": null, - "software.framework.jetpack": "6.0", - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.1", - "software.framework.vllm": null, - "software.operating_system": "Jetson r36.3.0 L4T", - "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "ConnectTechInc", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": null, - "system.accelerator.memory_config": "LPDDR5", - "system.accelerator.name": "NVIDIA Jetson AGX Orin 64G", - "system.accelerator.total_count": 1, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 12, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "12-core ARM Cortex-A78AE CPU", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "ARM", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 64.0, - "system.memory.configuration": "64GB 256-bit LPDDR5", - "system.name": "NVIDIA Jetson AGX Orin 64G (TensorRT) + CTI Forge Carrier (AGX201)", - "system.number_of_nodes": 1, - "system.type": "edge" - }, - { - "metrics.accuracy": "ROUGE1: 43.0343 ROUGE2: 20.0912 ROUGEL: 29.9254 GEN_LEN: 4102194", - "metrics.result": 7191.41, - "metrics.result_per_accelerator": 1797.8525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.697946695208735 exact_match: 90.06889808074122 TOKENS_PER_SAMPLE: 644.8", + "metrics.result": 557.097, + "metrics.result_per_accelerator": 69.637125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe", - "system.interconnect.accelerator_host": "PCIe", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB 4400 MT/s", - "system.name": "Dell PowerEdge R760xa (4xH100 NVL, TensorRT)", + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0347 ROUGE2: 20.0913 ROUGEL: 29.9262 GEN_LEN: 4102176", - "metrics.result": 6337.64, - "metrics.result_per_accelerator": 1584.41, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.678324128995275 exact_match: 90.0528821972204 TOKENS_PER_SAMPLE: 650.3", + "metrics.result": 291.183, + "metrics.result_per_accelerator": 36.397875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe", - "system.interconnect.accelerator_host": "PCIe", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB 4400 MT/s", - "system.name": "Dell PowerEdge R760xa (4xH100 NVL, TensorRT)", + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1259 ROUGE2: 20.1653 ROUGEL: 30.0252 GEN_LEN: 4101368", - "metrics.result": 3317.08, - "metrics.result_per_accelerator": 1658.54, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34809.3, + "metrics.result_per_accelerator": 4351.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe", - "system.interconnect.accelerator_host": "PCIe", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB 4400 MT/s", - "system.name": "Dell PowerEdge R760 (2xH100_PCIe_80GB, TensorRT)", + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1239 ROUGE2: 20.1642 ROUGEL: 30.0234 GEN_LEN: 4101242", - "metrics.result": 2759.33, - "metrics.result_per_accelerator": 1379.665, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5018 ROUGE2: 22.0993 ROUGEL: 28.7329 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 19418.5, + "metrics.result_per_accelerator": 2427.3125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe", - "system.interconnect.accelerator_host": "PCIe", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB 4400 MT/s", - "system.name": "Dell PowerEdge R760 (2xH100_PCIe_80GB, TensorRT)", + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1259 ROUGE2: 20.1653 ROUGEL: 30.0252 GEN_LEN: 4101368", - "metrics.result": 3317.08, - "metrics.result_per_accelerator": 1658.54, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5468 ROUGE2: 22.134 ROUGEL: 28.7841 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34807.1, + "metrics.result_per_accelerator": 4350.8875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "ASUSTeK", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe", - "system.interconnect.accelerator_host": "PCIe", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB 4400 MT/s", - "system.name": "Dell PowerEdge R760 (2xH100_PCIe_80GB, TensorRT)", + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1239 ROUGE2: 20.1642 ROUGEL: 30.0234 GEN_LEN: 4101242", - "metrics.result": 2759.33, - "metrics.result_per_accelerator": 1379.665, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5621 ROUGE2: 22.1419 ROUGEL: 28.7948 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 33037.2, + "metrics.result_per_accelerator": 4129.65, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "ASUSTeK", "submission.scenario": "Server", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM2e", - "system.accelerator.name": "NVIDIA H100-PCIe-80GB", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 120, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8558", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe", - "system.interconnect.accelerator_host": "PCIe", + "system.interconnect.accelerator": null, + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB 4400 MT/s", - "system.name": "Dell PowerEdge R760 (2xH100_PCIe_80GB, TensorRT)", + "system.memory.configuration": "32x 64GB ", + "system.name": "ASUSTeK ESC N8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 9966.79, - "metrics.result_per_accelerator": 2491.6975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 19936.5, + "metrics.result_per_accelerator": 2492.0625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Broadcom_Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.name": "Virtualized NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "32x 98GB MTC40F204WS1RC48BB1MHCC", + "system.name": " SYS-821GE-TNRT (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0994 ROUGE2: 20.1104 ROUGEL: 29.9725 GEN_LEN: 4114148", - "metrics.result": 9990.24, - "metrics.result_per_accelerator": 2497.56, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 19318.9, + "metrics.result_per_accelerator": 2414.8625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Broadcom_Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.name": "Virtualized NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "32x 98GB MTC40F204WS1RC48BB1MHCC", + "system.name": " SYS-821GE-TNRT (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 9966.79, - "metrics.result_per_accelerator": 2491.6975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 19936.5, + "metrics.result_per_accelerator": 2492.0625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Broadcom_Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.name": "Virtualized NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "32x 98GB MTC40F204WS1RC48BB1MHCC", + "system.name": " SYS-821GE-TNRT (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0994 ROUGE2: 20.1104 ROUGEL: 29.9725 GEN_LEN: 4114148", - "metrics.result": 9990.24, - "metrics.result_per_accelerator": 2497.56, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 19318.9, + "metrics.result_per_accelerator": 2414.8625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "Broadcom_Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.name": "Virtualized NVIDIA H100-SXM-80GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "32x 98GB MTC40F204WS1RC48BB1MHCC", + "system.name": " SYS-821GE-TNRT (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5354 ROUGE2: 22.0994 ROUGEL: 28.7324 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 10699.6, - "metrics.result_per_accelerator": 2674.9, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.3988 ROUGE2: 22.039 ROUGEL: 28.6453 TOKENS_PER_SAMPLE: 295.3", + "metrics.result": 33251.3, + "metrics.result_per_accelerator": 4156.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "ROCm 6.3.3", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.rocm": "6.3.3", "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 2048.0, + "system.accelerator.memory_config": "HBM3E", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", + "system.cpu.caches": "512 MB", + "system.cpu.core_count": 16, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 5.0, + "system.cpu.model": "2xAMD EPY 9175F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB Samsung M321RAJA0MB2-CCPPC", + "system.name": "G893-ZX1-AAX2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5352 ROUGE2: 22.0997 ROUGEL: 28.7319 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 9522.2, - "metrics.result_per_accelerator": 2380.55, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4157 ROUGE2: 22.0162 ROUGEL: 28.6494 TOKENS_PER_SAMPLE: 296.1", + "metrics.result": 29094.3, + "metrics.result_per_accelerator": 3636.7875, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "ROCm 6.3.3", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.rocm": "6.3.3", "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 2048.0, + "system.accelerator.memory_config": "HBM3E", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", + "system.cpu.caches": "512 MB", + "system.cpu.core_count": 16, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 5.0, + "system.cpu.model": "2xAMD EPY 9175F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB Samsung M321RAJA0MB2-CCPPC", + "system.name": "G893-ZX1-AAX2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5354 ROUGE2: 22.0994 ROUGEL: 28.7324 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 10699.6, - "metrics.result_per_accelerator": 2674.9, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.3988 ROUGE2: 22.039 ROUGEL: 28.6453 TOKENS_PER_SAMPLE: 295.3", + "metrics.result": 33251.3, + "metrics.result_per_accelerator": 4156.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "ROCm 6.3.3", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.rocm": "6.3.3", "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 2048.0, + "system.accelerator.memory_config": "HBM3E", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", + "system.cpu.caches": "512 MB", + "system.cpu.core_count": 16, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 5.0, + "system.cpu.model": "2xAMD EPY 9175F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB Samsung M321RAJA0MB2-CCPPC", + "system.name": "G893-ZX1-AAX2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5352 ROUGE2: 22.0997 ROUGEL: 28.7319 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 9522.2, - "metrics.result_per_accelerator": 2380.55, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.4157 ROUGE2: 22.0162 ROUGEL: 28.6494 TOKENS_PER_SAMPLE: 296.1", + "metrics.result": 29094.3, + "metrics.result_per_accelerator": 3636.7875, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "ROCm 6.3.3", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.rocm": "6.3.3", "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.5", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 2048.0, + "system.accelerator.memory_config": "HBM3E", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", + "system.cpu.caches": "512 MB", + "system.cpu.core_count": 16, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 5.0, + "system.cpu.model": "2xAMD EPY 9175F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB Samsung M321RAJA0MB2-CCPPC", + "system.name": "G893-ZX1-AAX2", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3721 ROUGE2: 23.2543 ROUGEL: 30.3259 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.12 mbxp_accuracy: 59.84", - "metrics.result": 26824.4, - "metrics.result_per_accelerator": 6706.1, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34889.5, + "metrics.result_per_accelerator": 4361.1875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3594 ROUGE2: 23.2482 ROUGEL: 30.3186 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.08 mbxp_accuracy: 59.9", - "metrics.result": 25707.0, - "metrics.result_per_accelerator": 6426.75, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5553 ROUGE2: 22.1419 ROUGEL: 28.79 TOKENS_PER_SAMPLE: 292.5", + "metrics.result": 33054.0, + "metrics.result_per_accelerator": 4131.75, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE8640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1787 ROUGEL: 28.7997 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 10594.1, - "metrics.result_per_accelerator": 2648.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21450.8, + "metrics.result_per_accelerator": 2681.35, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE9640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6141 ROUGE2: 22.1791 ROUGEL: 28.7992 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 8937.82, - "metrics.result_per_accelerator": 2234.455, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21544.3, + "metrics.result_per_accelerator": 2693.0375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE9640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1787 ROUGEL: 28.7997 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 10594.1, - "metrics.result_per_accelerator": 2648.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21450.8, + "metrics.result_per_accelerator": 2681.35, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE9640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6141 ROUGE2: 22.1791 ROUGEL: 28.7992 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 8937.82, - "metrics.result_per_accelerator": 2234.455, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21544.3, + "metrics.result_per_accelerator": 2693.0375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.6", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB DDR5", - "system.name": "Dell PowerEdge XE9640 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 24086.7, - "metrics.result_per_accelerator": 3010.8375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34986.7, + "metrics.result_per_accelerator": 4373.3375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6146 ROUGE2: 22.1791 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 21589.2, - "metrics.result_per_accelerator": 2698.65, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5085 ROUGE2: 22.1058 ROUGEL: 28.7374 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 20234.5, + "metrics.result_per_accelerator": 2529.3125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 24086.7, - "metrics.result_per_accelerator": 3010.8375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.6350817800776 exact_match: 90.03330906684316 TOKENS_PER_SAMPLE: 645.1", + "metrics.result": 558.526, + "metrics.result_per_accelerator": 69.81575, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6146 ROUGE2: 22.1791 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 21589.2, - "metrics.result_per_accelerator": 2698.65, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.63593824193972 exact_match: 89.99383520847121 TOKENS_PER_SAMPLE: 644.8", + "metrics.result": 288.601, + "metrics.result_per_accelerator": 36.075125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 20238.4, - "metrics.result_per_accelerator": 2529.8, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34986.7, + "metrics.result_per_accelerator": 4373.3375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -27768,45 +29678,48 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1012 ROUGE2: 20.113 ROUGEL: 29.9749 GEN_LEN: 4114492", - "metrics.result": 20139.0, - "metrics.result_per_accelerator": 2517.375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5085 ROUGE2: 22.1058 ROUGEL: 28.7374 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 20234.5, + "metrics.result_per_accelerator": 2529.3125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -27814,45 +29727,48 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 20238.4, - "metrics.result_per_accelerator": 2529.8, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4064 ROUGE2: 23.1865 ROUGEL: 30.2546 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.12", + "metrics.result": 62386.2, + "metrics.result_per_accelerator": 7798.275, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -27860,45 +29776,48 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1012 ROUGE2: 20.113 ROUGEL: 29.9749 GEN_LEN: 4114492", - "metrics.result": 20139.0, - "metrics.result_per_accelerator": 2517.375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4078 ROUGE2: 23.2531 ROUGEL: 30.3504 TOKENS_PER_SAMPLE: 146.3 gsm8k_accuracy: 74.04 mbxp_accuracy: 59.78", + "metrics.result": 60205.0, + "metrics.result_per_accelerator": 7525.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -27906,45 +29825,48 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5883 ROUGE2: 22.1723 ROUGEL: 28.7904 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 32124.3, - "metrics.result_per_accelerator": 4015.5375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34889.5, + "metrics.result_per_accelerator": 4361.1875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -27952,45 +29874,48 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5856 ROUGE2: 22.1708 ROUGEL: 28.7886 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 29739.9, - "metrics.result_per_accelerator": 3717.4875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5553 ROUGE2: 22.1419 ROUGEL: 28.79 TOKENS_PER_SAMPLE: 292.5", + "metrics.result": 33054.0, + "metrics.result_per_accelerator": 4131.75, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "GigaComputing", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -27998,3265 +29923,3478 @@ "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.caches": "105 MB", + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.frequency": 3.8, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", + "system.name": "G893-SD1", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5883 ROUGE2: 22.1723 ROUGEL: 28.7904 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 32124.3, - "metrics.result_per_accelerator": 4015.5375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 22.02784417487302 exact_match: 90.10202845797485 TOKENS_PER_SAMPLE: 655.4", + "metrics.result": 801.906, + "metrics.result_per_accelerator": 200.4765, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "coreweave", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 186.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GB200", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "for each 4-GPU node: 1.3TB LPDDR5x", + "system.name": " (4 x GB200-186GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5856 ROUGE2: 22.1708 ROUGEL: 28.7886 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 29739.9, - "metrics.result_per_accelerator": 3717.4875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.989455826099128 exact_match: 90.08294837855725 TOKENS_PER_SAMPLE: 654.2", + "metrics.result": 522.12, + "metrics.result_per_accelerator": 130.53, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "coreweave", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 186.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.name": "NVIDIA GB200", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 52, + "system.cpu.core_count": 144, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8470", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x H200-SXM-141GB, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 960.0, + "system.memory.configuration": "for each 4-GPU node: 1.3TB LPDDR5x", + "system.name": " (4 x GB200-186GB_aarch64, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5205 ROUGE2: 22.0831 ROUGEL: 28.7052 TOKENS_PER_SAMPLE: 293.6", - "metrics.result": 22677.6, - "metrics.result_per_accelerator": 2834.7, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34403.5, + "metrics.result_per_accelerator": 4300.4375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.0", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.0", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.4", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "coreweave", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 40, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x MI300X_192GB, vLLM)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4783 ROUGE2: 22.0575 ROUGEL: 28.6859 TOKENS_PER_SAMPLE: 293.8", - "metrics.result": 19886.1, - "metrics.result_per_accelerator": 2485.7625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5641 ROUGE2: 22.1454 ROUGEL: 28.794 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 33066.5, + "metrics.result_per_accelerator": 4133.3125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.0", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.0", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.4", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "coreweave", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 40, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x MI300X_192GB, vLLM)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5205 ROUGE2: 22.0831 ROUGEL: 28.7052 TOKENS_PER_SAMPLE: 293.6", - "metrics.result": 22677.6, - "metrics.result_per_accelerator": 2834.7, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34403.5, + "metrics.result_per_accelerator": 4300.4375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.0", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.0", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.4", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "coreweave", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 40, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x MI300X_192GB, vLLM)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4783 ROUGE2: 22.0575 ROUGEL: 28.6859 TOKENS_PER_SAMPLE: 293.8", - "metrics.result": 19886.1, - "metrics.result_per_accelerator": 2485.7625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5641 ROUGE2: 22.1454 ROUGEL: 28.794 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 33066.5, + "metrics.result_per_accelerator": 4133.3125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "vLLM 0.4.3+rocm614, PyTorch 2.3.0, ROCm 6.1.0", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": "2.3.0", - "software.framework.rocm": "6.1.0", + "software.framework.pytorch": null, + "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.4.3", - "software.operating_system": "Ubuntu 22.04.4", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Dell", + "submission.organization": "coreweave", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 192.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "AMD MI300X-NPS1-SPX-192GB-750W", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, - "system.accelerator.vendor": "AMD", + "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 40, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8460Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "Dell PowerEdge XE9680 (8x MI300X_192GB, vLLM)", + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1387 ROUGE2: 20.1794 ROUGEL: 30.0469 GEN_LEN: 4112577", - "metrics.result": 6911.78, - "metrics.result_per_accelerator": 863.9725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5825 ROUGE2: 22.1545 ROUGEL: 28.776 TOKENS_PER_SAMPLE: 291.8", + "metrics.result": 30899.7, + "metrics.result_per_accelerator": 3862.4625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.4, - "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9554 64-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", - "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", + "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1388 ROUGE2: 20.1792 ROUGEL: 30.0469 GEN_LEN: 4112636", - "metrics.result": 6903.45, - "metrics.result_per_accelerator": 862.93125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6229 ROUGE2: 22.2108 ROUGEL: 28.8257 TOKENS_PER_SAMPLE: 291.6", + "metrics.result": 26599.4, + "metrics.result_per_accelerator": 3324.925, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.4, - "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9554 64-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", - "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", + "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1387 ROUGE2: 20.1794 ROUGEL: 30.0469 GEN_LEN: 4112577", - "metrics.result": 6911.78, - "metrics.result_per_accelerator": 863.9725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20307.7, + "metrics.result_per_accelerator": 2538.4625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.4, - "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9554 64-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", - "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", + "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1388 ROUGE2: 20.1792 ROUGEL: 30.0469 GEN_LEN: 4112636", - "metrics.result": 6903.45, - "metrics.result_per_accelerator": 862.93125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20706.7, + "metrics.result_per_accelerator": 2588.3375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.4, - "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9554 64-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", - "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", + "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5205 ROUGE2: 22.0796 ROUGEL: 28.6977 TOKENS_PER_SAMPLE: 292.9", - "metrics.result": 3717.74, - "metrics.result_per_accelerator": 464.7175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20307.7, + "metrics.result_per_accelerator": 2538.4625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.4, - "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9554 64-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", - "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", + "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5132 ROUGE2: 22.0915 ROUGEL: 28.7055 TOKENS_PER_SAMPLE: 293.0", - "metrics.result": 3218.55, - "metrics.result_per_accelerator": 402.31875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20706.7, + "metrics.result_per_accelerator": 2588.3375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.4, - "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9554 64-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", - "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", + "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5205 ROUGE2: 22.0796 ROUGEL: 28.6977 TOKENS_PER_SAMPLE: 292.9", - "metrics.result": 3717.74, - "metrics.result_per_accelerator": 464.7175, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.5315 ROUGE2: 23.3511 ROUGEL: 30.4188 TOKENS_PER_SAMPLE: 144.9 gsm8k_accuracy: 73.28 mbxp_accuracy: 60.06", + "metrics.result": 52724.4, + "metrics.result_per_accelerator": 6590.55, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.4, - "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9554 64-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", - "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", + "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5132 ROUGE2: 22.0915 ROUGEL: 28.7055 TOKENS_PER_SAMPLE: 293.0", - "metrics.result": 3218.55, - "metrics.result_per_accelerator": 402.31875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.5484 ROUGE2: 23.3628 ROUGEL: 30.443 TOKENS_PER_SAMPLE: 146.5 gsm8k_accuracy: 72.96 mbxp_accuracy: 59.6", + "metrics.result": 53299.3, + "metrics.result_per_accelerator": 6662.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d:3MiB, L1i:2MiB, L2:128MiB, L3:120MiB", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.4, - "system.cpu.model": "Intel(R) Xeon(R) Gold 6454S", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9554 64-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ", - "system.name": "PRIMERGY CDI (8x L40S, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", + "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 9960.74, - "metrics.result_per_accelerator": 2490.185, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5825 ROUGE2: 22.1545 ROUGEL: 28.776 TOKENS_PER_SAMPLE: 291.8", + "metrics.result": 30899.7, + "metrics.result_per_accelerator": 3862.4625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9554 64-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "32x 32GB DDR5", - "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", + "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0991 ROUGE2: 20.1109 ROUGEL: 29.9733 GEN_LEN: 4114114", - "metrics.result": 9602.79, - "metrics.result_per_accelerator": 2400.6975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6229 ROUGE2: 22.2108 ROUGEL: 28.8257 TOKENS_PER_SAMPLE: 291.6", + "metrics.result": 26599.4, + "metrics.result_per_accelerator": 3324.925, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9554 64-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "32x 32GB DDR5", - "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB HMCGM4MGBRB240N", + "system.name": "Cisco UCS C885A M8 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 9960.74, - "metrics.result_per_accelerator": 2490.185, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0554 ROUGE2: 20.0943 ROUGEL: 29.9455 GEN_LEN: 4157784", + "metrics.result": 1754.97, + "metrics.result_per_accelerator": 877.485, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 64.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9224 24-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "32x 32GB DDR5", - "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB ", + "system.name": "UCS X215 M8 (2x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0991 ROUGE2: 20.1109 ROUGEL: 29.9733 GEN_LEN: 4114114", - "metrics.result": 9602.79, - "metrics.result_per_accelerator": 2400.6975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0249 ROUGE2: 20.0689 ROUGEL: 29.9271 GEN_LEN: 4153891", + "metrics.result": 1745.78, + "metrics.result_per_accelerator": 872.89, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 64.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 64, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9224 24-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1024.0, - "system.memory.configuration": "32x 32GB DDR5", - "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "24x 64GB ", + "system.name": "UCS X215 M8 (2x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1787 ROUGEL: 28.7997 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 10133.3, - "metrics.result_per_accelerator": 2533.325, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.514 ROUGE2: 22.1255 ROUGEL: 28.7527 TOKENS_PER_SAMPLE: 291.1", + "metrics.result": 3741.32, + "metrics.result_per_accelerator": 1870.66, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "NVL", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 24, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9224", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "32x 32GB DDR5", - "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB 5600 DIMMs", + "system.name": "Cisco UCS C245 M8 (2x H100NVL-PCIe-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6146 ROUGE2: 22.1791 ROUGEL: 28.8001 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 9493.99, - "metrics.result_per_accelerator": 2373.4975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5103 ROUGE2: 22.1066 ROUGEL: 28.7387 TOKENS_PER_SAMPLE: 292.2", + "metrics.result": 3337.55, + "metrics.result_per_accelerator": 1668.775, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "NVL", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 24, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9224", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "32x 32GB DDR5", - "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB 5600 DIMMs", + "system.name": "Cisco UCS C245 M8 (2x H100NVL-PCIe-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1787 ROUGEL: 28.7997 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 10133.3, - "metrics.result_per_accelerator": 2533.325, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 3789.42, + "metrics.result_per_accelerator": 1894.71, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "NVL", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 24, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9224", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "32x 32GB DDR5", - "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB 5600 DIMMs", + "system.name": "Cisco UCS C245 M8 (2x H100NVL-PCIe-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6146 ROUGE2: 22.1791 ROUGEL: 28.8001 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 9493.99, - "metrics.result_per_accelerator": 2373.4975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 3305.27, + "metrics.result_per_accelerator": 1652.635, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "NVL", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 24, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9224", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "32x 32GB DDR5", - "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB 5600 DIMMs", + "system.name": "Cisco UCS C245 M8 (2x H100NVL-PCIe-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3504 ROUGE2: 23.2436 ROUGEL: 30.3147 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.08 mbxp_accuracy: 59.9", - "metrics.result": 26641.6, - "metrics.result_per_accelerator": 6660.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.514 ROUGE2: 22.1255 ROUGEL: 28.7527 TOKENS_PER_SAMPLE: 291.1", + "metrics.result": 3879.95, + "metrics.result_per_accelerator": 1939.975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": null, + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9554", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "32x 32GB DDR5", - "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB 5600 DIMMs", + "system.name": "Cisco UCS X215c M8 with X440P PCIe node (2x H100NVL-PCIe-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3591 ROUGE2: 23.2517 ROUGEL: 30.3173 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", - "metrics.result": 25358.5, - "metrics.result_per_accelerator": 6339.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5111 ROUGE2: 22.1063 ROUGEL: 28.7376 TOKENS_PER_SAMPLE: 292.2", + "metrics.result": 3337.8, + "metrics.result_per_accelerator": 1668.9, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Fujitsu", + "submission.organization": "Cisco", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": null, + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9554", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1024.0, - "system.memory.configuration": "32x 32GB DDR5", - "system.name": "GX2560M7_H100_SXM_80GBx4 (4x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 64GB 5600 DIMMs", + "system.name": "Cisco UCS X215c M8 with X440P PCIe node (2x H100NVL-PCIe-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 20041.3, - "metrics.result_per_accelerator": 2505.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 3743.36, + "metrics.result_per_accelerator": 1871.68, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Cisco", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": null, + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9554", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "GIGABYTE G593-SD1", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB 5600 DIMMs", + "system.name": "Cisco UCS X215c M8 with X440P PCIe node (2x H100NVL-PCIe-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1004 ROUGE2: 20.1119 ROUGEL: 29.9737 GEN_LEN: 4114146", - "metrics.result": 19250.4, - "metrics.result_per_accelerator": 2406.3, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 3701.78, + "metrics.result_per_accelerator": 1850.89, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Cisco", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 2, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": null, + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 2, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9554", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "GIGABYTE G593-SD1", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB 5600 DIMMs", + "system.name": "Cisco UCS X215c M8 with X440P PCIe node (2x H100NVL-PCIe-94GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 20041.3, - "metrics.result_per_accelerator": 2505.1625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34081.1, + "metrics.result_per_accelerator": 4260.1375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Cisco", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", + "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "GIGABYTE G593-SD1", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", + "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1004 ROUGE2: 20.1119 ROUGEL: 29.9737 GEN_LEN: 4114146", - "metrics.result": 19250.4, - "metrics.result_per_accelerator": 2406.3, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5122 ROUGE2: 22.1039 ROUGEL: 28.7358 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 30420.2, + "metrics.result_per_accelerator": 3802.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Cisco", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", + "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "GIGABYTE G593-SD1", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", + "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5884 ROUGE2: 22.1723 ROUGEL: 28.7905 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 31263.8, - "metrics.result_per_accelerator": 3907.975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21557.8, + "metrics.result_per_accelerator": 2694.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Cisco", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", + "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "GIGABYTE G593-SD1", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", + "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5891 ROUGE2: 22.1726 ROUGEL: 28.7906 TOKENS_PER_SAMPLE: 291.6", - "metrics.result": 29715.3, - "metrics.result_per_accelerator": 3714.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21813.4, + "metrics.result_per_accelerator": 2726.675, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Cisco", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", + "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "GIGABYTE G593-SD1", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", + "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5884 ROUGE2: 22.1723 ROUGEL: 28.7905 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 31263.8, - "metrics.result_per_accelerator": 3907.975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21557.8, + "metrics.result_per_accelerator": 2694.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Cisco", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", + "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "GIGABYTE G593-SD1", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", + "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5891 ROUGE2: 22.1726 ROUGEL: 28.7906 TOKENS_PER_SAMPLE: 291.6", - "metrics.result": 29715.3, - "metrics.result_per_accelerator": 3714.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21813.4, + "metrics.result_per_accelerator": 2726.675, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Cisco", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", + "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "GIGABYTE G593-SD1", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", + "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3568 ROUGE2: 23.2496 ROUGEL: 30.3171 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", - "metrics.result": 59334.8, - "metrics.result_per_accelerator": 7416.85, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 35102.5, + "metrics.result_per_accelerator": 4387.8125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Cisco", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", + "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "GIGABYTE G593-SD1", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", + "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3551 ROUGE2: 23.245 ROUGEL: 30.3156 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", - "metrics.result": 57174.1, - "metrics.result_per_accelerator": 7146.7625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5023 ROUGE2: 22.0975 ROUGEL: 28.7294 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 19437.4, + "metrics.result_per_accelerator": 2429.675, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "GigaComputing", + "submission.organization": "Cisco", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", + "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480+", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5", - "system.name": "GIGABYTE G593-SD1", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", + "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6156 ROUGE2: 22.1792 ROUGEL: 28.8002 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 24133.7, - "metrics.result_per_accelerator": 3016.7125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.792712875044835 exact_match: 90.00425876902713 TOKENS_PER_SAMPLE: 660.9", + "metrics.result": 573.573, + "metrics.result_per_accelerator": 71.696625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "Cisco", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.7, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8481C", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", + "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6146 ROUGE2: 22.1791 ROUGEL: 28.7999 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 21588.3, - "metrics.result_per_accelerator": 2698.5375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.698924776508495 exact_match: 90.06501323626736 TOKENS_PER_SAMPLE: 657.5", + "metrics.result": 294.275, + "metrics.result_per_accelerator": 36.784375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "Cisco", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.7, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8481C", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", + "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6156 ROUGE2: 22.1792 ROUGEL: 28.8002 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 24133.7, - "metrics.result_per_accelerator": 3016.7125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34081.1, + "metrics.result_per_accelerator": 4260.1375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "Cisco", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.7, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8481C", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", + "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6146 ROUGE2: 22.1791 ROUGEL: 28.7999 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 21588.3, - "metrics.result_per_accelerator": 2698.5375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5122 ROUGE2: 22.1039 ROUGEL: 28.7358 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 30420.2, + "metrics.result_per_accelerator": 3802.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "Cisco", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 128, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.7, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8481C", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB MTC40F2046S1RC48BA1", + "system.name": "Cisco UCS C885A M8 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3671 ROUGE2: 23.2507 ROUGEL: 30.3151 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", - "metrics.result": 52524.4, - "metrics.result_per_accelerator": 6565.55, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 35385.1, + "metrics.result_per_accelerator": 4423.1375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "Lambda", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.7, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8481C", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3561 ROUGE2: 23.2477 ROUGEL: 30.3155 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.08 mbxp_accuracy: 59.88", - "metrics.result": 50798.8, - "metrics.result_per_accelerator": 6349.85, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5554 ROUGE2: 22.138 ROUGEL: 28.7885 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 33045.3, + "metrics.result_per_accelerator": 4130.6625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Google", + "submission.organization": "Lambda", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 56, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 2.7, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8481C", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 2632.76, - "metrics.result_per_accelerator": 2632.76, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21448.4, + "metrics.result_per_accelerator": 2681.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 24.04", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Lambda", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1002 ROUGE2: 20.1114 ROUGEL: 29.9742 GEN_LEN: 4114259", - "metrics.result": 2512.89, - "metrics.result_per_accelerator": 2512.89, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21543.9, + "metrics.result_per_accelerator": 2692.9875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 24.04", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Lambda", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 2632.76, - "metrics.result_per_accelerator": 2632.76, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21448.4, + "metrics.result_per_accelerator": 2681.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 24.04", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Lambda", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1002 ROUGE2: 20.1114 ROUGEL: 29.9742 GEN_LEN: 4114259", - "metrics.result": 2512.89, - "metrics.result_per_accelerator": 2512.89, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21543.9, + "metrics.result_per_accelerator": 2692.9875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 24.04", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Lambda", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 4084.3, - "metrics.result_per_accelerator": 4084.3, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 35409.4, + "metrics.result_per_accelerator": 4426.175, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 24.04", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Lambda", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1577 ROUGEL: 28.7741 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 3884.16, - "metrics.result_per_accelerator": 3884.16, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5019 ROUGE2: 22.0977 ROUGEL: 28.7293 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 19381.0, + "metrics.result_per_accelerator": 2422.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 24.04", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Lambda", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 4084.3, - "metrics.result_per_accelerator": 4084.3, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 35409.4, + "metrics.result_per_accelerator": 4426.175, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 24.04", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Lambda", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1577 ROUGEL: 28.7741 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 3884.16, - "metrics.result_per_accelerator": 3884.16, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5019 ROUGE2: 22.0977 ROUGEL: 28.7293 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 19381.0, + "metrics.result_per_accelerator": 2422.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 24.04", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Lambda", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3565 ROUGE2: 23.2488 ROUGEL: 30.3157 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", - "metrics.result": 8063.02, - "metrics.result_per_accelerator": 8063.02, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 35385.1, + "metrics.result_per_accelerator": 4423.1375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 24.04", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Lambda", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3563 ROUGE2: 23.2486 ROUGEL: 30.3162 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.94", - "metrics.result": 7450.72, - "metrics.result_per_accelerator": 7450.72, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5554 ROUGE2: 22.138 ROUGEL: 28.7885 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 33045.3, + "metrics.result_per_accelerator": 4130.6625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 24.04", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Lambda", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 56, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "HPE ProLiant Compute DL384 Gen12 (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", + "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0343 ROUGE2: 20.0912 ROUGEL: 29.9254 GEN_LEN: 4102194", - "metrics.result": 19751.9, - "metrics.result_per_accelerator": 2468.9875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", + "metrics.result": 97416.9, + "metrics.result_per_accelerator": 12177.1125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0345 ROUGE2: 20.0914 ROUGEL: 29.9253 GEN_LEN: 4102160", - "metrics.result": 19502.6, - "metrics.result_per_accelerator": 2437.825, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7857 ROUGE2: 22.3844 ROUGEL: 29.1435 TOKENS_PER_SAMPLE: 273.8", + "metrics.result": 97552.1, + "metrics.result_per_accelerator": 12194.0125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.535 ROUGE2: 22.0991 ROUGEL: 28.7319 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 24528.3, - "metrics.result_per_accelerator": 3066.0375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.8033 ROUGE2: 22.3646 ROUGEL: 29.1684 TOKENS_PER_SAMPLE: 271.8", + "metrics.result": 92689.2, + "metrics.result_per_accelerator": 11586.15, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5353 ROUGE2: 22.0997 ROUGEL: 28.7319 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 23132.8, - "metrics.result_per_accelerator": 2891.6, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.8117 ROUGE2: 22.3683 ROUGEL: 29.1731 TOKENS_PER_SAMPLE: 272.7", + "metrics.result": 62265.7, + "metrics.result_per_accelerator": 7783.2125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.535 ROUGE2: 22.0991 ROUGEL: 28.7319 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 24424.7, - "metrics.result_per_accelerator": 3053.0875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 22.092192082812325 exact_match: 90.10742885506286 TOKENS_PER_SAMPLE: 654.1", + "metrics.result": 1521.74, + "metrics.result_per_accelerator": 190.2175, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5353 ROUGE2: 22.0997 ROUGEL: 28.732 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 23144.0, - "metrics.result_per_accelerator": 2893.0, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 22.071095668630047 exact_match: 90.1108107213766 TOKENS_PER_SAMPLE: 655.5", + "metrics.result": 1080.31, + "metrics.result_per_accelerator": 135.03875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3566 ROUGE2: 23.2487 ROUGEL: 30.3174 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", - "metrics.result": 52395.1, - "metrics.result_per_accelerator": 6549.3875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.8033 ROUGE2: 22.3646 ROUGEL: 29.1684 TOKENS_PER_SAMPLE: 271.8", + "metrics.result": 92689.2, + "metrics.result_per_accelerator": 11586.15, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3565 ROUGE2: 23.249 ROUGEL: 30.3137 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", - "metrics.result": 50798.5, - "metrics.result_per_accelerator": 6349.8125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.8117 ROUGE2: 22.3683 ROUGEL: 29.1731 TOKENS_PER_SAMPLE: 272.7", + "metrics.result": 62265.7, + "metrics.result_per_accelerator": 7783.2125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6113 ROUGE2: 22.1451 ROUGEL: 28.7779 TOKENS_PER_SAMPLE: 290.3", - "metrics.result": 8858.78, - "metrics.result_per_accelerator": 2214.695, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.5759 ROUGE2: 23.2321 ROUGEL: 30.3362 TOKENS_PER_SAMPLE: 144.4 gsm8k_accuracy: 73.38 mbxp_accuracy: 60.3", + "metrics.result": 125534.0, + "metrics.result_per_accelerator": 15691.75, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB, L1i: 2 MiB, L2: 128 MiB (64 instances), L3: 320 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB", - "system.name": "HPE ProLiant DL380a Gen11 (4x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6118 ROUGE2: 22.1456 ROUGEL: 28.7778 TOKENS_PER_SAMPLE: 290.3", - "metrics.result": 6927.8, - "metrics.result_per_accelerator": 1731.95, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4923 ROUGE2: 23.2325 ROUGEL: 30.3051 TOKENS_PER_SAMPLE: 146.6 gsm8k_accuracy: 73.02 mbxp_accuracy: 59.72", + "metrics.result": 128961.0, + "metrics.result_per_accelerator": 16120.125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB, L1i: 2 MiB, L2: 128 MiB (64 instances), L3: 320 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB", - "system.name": "HPE ProLiant DL380a Gen11 (4x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6113 ROUGE2: 22.1451 ROUGEL: 28.7779 TOKENS_PER_SAMPLE: 290.3", - "metrics.result": 8858.78, - "metrics.result_per_accelerator": 2214.695, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", + "metrics.result": 97416.9, + "metrics.result_per_accelerator": 12177.1125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB, L1i: 2 MiB, L2: 128 MiB (64 instances), L3: 320 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB", - "system.name": "HPE ProLiant DL380a Gen11 (4x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6118 ROUGE2: 22.1456 ROUGEL: 28.7778 TOKENS_PER_SAMPLE: 290.3", - "metrics.result": 6927.8, - "metrics.result_per_accelerator": 1731.95, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7857 ROUGE2: 22.3844 ROUGEL: 29.1435 TOKENS_PER_SAMPLE: 273.8", + "metrics.result": 97552.1, + "metrics.result_per_accelerator": 12194.0125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "HPE", + "submission.organization": "Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB, L1i: 2 MiB, L2: 128 MiB (64 instances), L3: 320 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB", - "system.name": "HPE ProLiant DL380a Gen11 (4x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 96GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-A21GE-NBRT (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6047 ROUGE2: 22.1268 ROUGEL: 28.7705 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 41672.4, - "metrics.result_per_accelerator": 2604.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34628.3, + "metrics.result_per_accelerator": 4328.5375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "JuniperNetworks", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 16, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 112, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", - "system.name": "2x8xH100-SXM-80GB", - "system.number_of_nodes": 2, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6052 ROUGE2: 22.1275 ROUGEL: 28.7709 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 41091.6, - "metrics.result_per_accelerator": 2568.225, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5615 ROUGE2: 22.1445 ROUGEL: 28.7927 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 33052.9, + "metrics.result_per_accelerator": 4131.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "JuniperNetworks", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 16, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 112, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", - "system.name": "2x8xH100-SXM-80GB", - "system.number_of_nodes": 2, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6047 ROUGE2: 22.1268 ROUGEL: 28.7705 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 41672.4, - "metrics.result_per_accelerator": 2604.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21304.3, + "metrics.result_per_accelerator": 2663.0375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "JuniperNetworks", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 16, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 112, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", - "system.name": "2x8xH100-SXM-80GB", - "system.number_of_nodes": 2, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6052 ROUGE2: 22.1275 ROUGEL: 28.7709 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 41091.6, - "metrics.result_per_accelerator": 2568.225, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21544.2, + "metrics.result_per_accelerator": 2693.025, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "JuniperNetworks", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 16, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 112, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", - "system.name": "2x8xH100-SXM-80GB", - "system.number_of_nodes": 2, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6048 ROUGE2: 22.1273 ROUGEL: 28.7697 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 82749.6, - "metrics.result_per_accelerator": 2585.925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21304.3, + "metrics.result_per_accelerator": 2663.0375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "JuniperNetworks", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 32, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 112, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", - "system.name": "4x8xH100-SXM-80GB", - "system.number_of_nodes": 4, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6056 ROUGE2: 22.1274 ROUGEL: 28.7709 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 82273.2, - "metrics.result_per_accelerator": 2571.0375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21544.2, + "metrics.result_per_accelerator": 2693.025, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "JuniperNetworks", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 32, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 112, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", - "system.name": "4x8xH100-SXM-80GB", - "system.number_of_nodes": 4, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6048 ROUGE2: 22.1273 ROUGEL: 28.7697 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 82749.6, - "metrics.result_per_accelerator": 2585.925, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5467 ROUGE2: 22.1337 ROUGEL: 28.7837 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34667.6, + "metrics.result_per_accelerator": 4333.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "JuniperNetworks", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 32, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 112, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", - "system.name": "4x8xH100-SXM-80GB", - "system.number_of_nodes": 4, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6056 ROUGE2: 22.1274 ROUGEL: 28.7709 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 82273.2, - "metrics.result_per_accelerator": 2571.0375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5006 ROUGE2: 22.1008 ROUGEL: 28.7314 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 19431.9, + "metrics.result_per_accelerator": 2428.9875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 20.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "JuniperNetworks", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 32, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 112, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 7763 64-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "PCIe", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32 slots / 64 GB each/ Total = 2 TB /DDR4 3200 MHz", - "system.name": "4x8xH100-SXM-80GB", - "system.number_of_nodes": 4, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", + "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0773 ROUGE2: 20.0872 ROUGEL: 29.9583 GEN_LEN: 4107356", - "metrics.result": 19859.2, - "metrics.result_per_accelerator": 2482.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5467 ROUGE2: 22.1337 ROUGEL: 28.7837 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34667.6, + "metrics.result_per_accelerator": 4333.45, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31268,41 +33406,44 @@ "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9454", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB HMCG94MEBRA121N", - "system.name": "ThinkSystem SR685a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0786 ROUGE2: 20.0893 ROUGEL: 29.9596 GEN_LEN: 4107309", - "metrics.result": 19716.0, - "metrics.result_per_accelerator": 2464.5, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5006 ROUGE2: 22.1008 ROUGEL: 28.7314 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 19431.9, + "metrics.result_per_accelerator": 2428.9875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31314,41 +33455,44 @@ "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9454", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB HMCG94MEBRA121N", - "system.name": "ThinkSystem SR685a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6249 ROUGE2: 22.1876 ROUGEL: 28.8076 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 31917.0, - "metrics.result_per_accelerator": 3989.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4086 ROUGE2: 23.1851 ROUGEL: 30.262 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.12", + "metrics.result": 61964.1, + "metrics.result_per_accelerator": 7745.5125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31360,41 +33504,44 @@ "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9454", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB HMCG94MEBRA121N", - "system.name": "ThinkSystem SR685a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6256 ROUGE2: 22.1883 ROUGEL: 28.808 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 30068.5, - "metrics.result_per_accelerator": 3758.5625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4991 ROUGE2: 23.3568 ROUGEL: 30.4444 TOKENS_PER_SAMPLE: 146.2 gsm8k_accuracy: 73.64 mbxp_accuracy: 60.14", + "metrics.result": 59627.3, + "metrics.result_per_accelerator": 7453.4125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31406,41 +33553,44 @@ "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9454", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB HMCG94MEBRA121N", - "system.name": "ThinkSystem SR685a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6249 ROUGE2: 22.1876 ROUGEL: 28.8076 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 31917.0, - "metrics.result_per_accelerator": 3989.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34628.3, + "metrics.result_per_accelerator": 4328.5375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31452,41 +33602,44 @@ "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9454", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB HMCG94MEBRA121N", - "system.name": "ThinkSystem SR685a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6256 ROUGE2: 22.1883 ROUGEL: 28.808 TOKENS_PER_SAMPLE: 290.9", - "metrics.result": 30068.5, - "metrics.result_per_accelerator": 3758.5625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5615 ROUGE2: 22.1445 ROUGEL: 28.7927 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 33052.9, + "metrics.result_per_accelerator": 4131.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31498,87 +33651,142 @@ "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9454", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB HMCG94MEBRA121N", - "system.name": "ThinkSystem SR685a V3(8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 4800MT/s @ 4400MT/s", + "system.name": "SYS-821GE-TNHR (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0266 ROUGE2: 20.1037 ROUGEL: 29.9635 GEN_LEN: 4086524", - "metrics.result": 1718.33, - "metrics.result_per_accelerator": 859.165, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4028 ROUGE2: 23.2139 ROUGEL: 30.3225 TOKENS_PER_SAMPLE: 145.1 gsm8k_accuracy: 73.14 mbxp_accuracy: 59.62", + "metrics.result": 7718.34, + "metrics.result_per_accelerator": 7718.34, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 9.3.0, CUDA 12.3", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.3", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "9.3.0", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 16, + "system.cpu.core_count": 72, "system.cpu.count_per_node": 1, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 8124P 16-Core Processor", + "system.cpu.model": "NVIDIA Grace CPU", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 128.0, - "system.memory.configuration": "4x 32GB M321R4GA3BB6-CQKDS", - "system.name": "ThinkEdge SE455 V3 (2x NVIDIA L40S, TensorRT)", + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "LPDDR5x", + "system.name": "ARS-111GL-NHR (1x GH200-96GB_aarch64)", "system.number_of_nodes": 1, - "system.type": "edge" + "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.102 ROUGE2: 20.1124 ROUGEL: 29.9753 GEN_LEN: 4114427", - "metrics.result": 20552.1, - "metrics.result_per_accelerator": 2569.0125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.5543 ROUGE2: 23.3316 ROUGEL: 30.4478 TOKENS_PER_SAMPLE: 146.1 gsm8k_accuracy: 73.72 mbxp_accuracy: 60.2", + "metrics.result": 7182.77, + "metrics.result_per_accelerator": 7182.77, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": null, "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 1, + "system.accelerator.memory_capacity": 96.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", + "system.accelerator.total_count": 1, + "system.accelerator.vendor": "NVIDIA", + "system.cpu.caches": null, + "system.cpu.core_count": 72, + "system.cpu.count_per_node": 1, + "system.cpu.frequency": null, + "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.vcpu_count": null, + "system.cpu.vendor": "NVIDIA", + "system.interconnect.accelerator": "InfiniBand", + "system.interconnect.accelerator_host": "NVLink", + "system.memory.capacity": 480.0, + "system.memory.configuration": "LPDDR5x", + "system.name": "ARS-111GL-NHR (1x GH200-96GB_aarch64)", + "system.number_of_nodes": 1, + "system.type": "datacenter" + }, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7836 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 35026.8, + "metrics.result_per_accelerator": 4378.35, + "metrics.units": "Tokens/s", + "model.architecture": "LLM", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", + "software.framework.ai": null, + "software.framework.cuda": "12.8", + "software.framework.cudnn": null, + "software.framework.jetpack": null, + "software.framework.pytorch": null, + "software.framework.rocm": null, + "software.framework.sima": null, + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", + "submission.availability": "available", + "submission.division": "closed", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31587,44 +33795,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.101 ROUGE2: 20.1122 ROUGEL: 29.9738 GEN_LEN: 4114231", - "metrics.result": 19785.0, - "metrics.result_per_accelerator": 2473.125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5599 ROUGE2: 22.1445 ROUGEL: 28.7952 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 33045.7, + "metrics.result_per_accelerator": 4130.7125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31633,44 +33844,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.102 ROUGE2: 20.1124 ROUGEL: 29.9753 GEN_LEN: 4114427", - "metrics.result": 20552.1, - "metrics.result_per_accelerator": 2569.0125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21477.4, + "metrics.result_per_accelerator": 2684.675, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31679,44 +33893,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.101 ROUGE2: 20.1122 ROUGEL: 29.9738 GEN_LEN: 4114231", - "metrics.result": 19785.0, - "metrics.result_per_accelerator": 2473.125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21757.3, + "metrics.result_per_accelerator": 2719.6625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31725,44 +33942,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5884 ROUGE2: 22.1723 ROUGEL: 28.7905 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 31973.7, - "metrics.result_per_accelerator": 3996.7125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21477.4, + "metrics.result_per_accelerator": 2684.675, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31771,44 +33991,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5866 ROUGE2: 22.1706 ROUGEL: 28.789 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 30438.7, - "metrics.result_per_accelerator": 3804.8375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21757.3, + "metrics.result_per_accelerator": 2719.6625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31817,44 +34040,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5884 ROUGE2: 22.1723 ROUGEL: 28.7905 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 31973.7, - "metrics.result_per_accelerator": 3996.7125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1334 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 35192.5, + "metrics.result_per_accelerator": 4399.0625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31863,44 +34089,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5866 ROUGE2: 22.1706 ROUGEL: 28.789 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 30438.7, - "metrics.result_per_accelerator": 3804.8375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5031 ROUGE2: 22.1063 ROUGEL: 28.7341 TOKENS_PER_SAMPLE: 292.5", + "metrics.result": 19413.2, + "metrics.result_per_accelerator": 2426.65, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -31909,2712 +34138,2889 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8568Y+", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "ThinkSystem SR680a V3 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0516 ROUGE2: 20.0942 ROUGEL: 29.9288 GEN_LEN: 4111570", - "metrics.result": 2593.65, - "metrics.result_per_accelerator": 864.5500000000001, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.50967980348964 exact_match: 90.01451025810722 TOKENS_PER_SAMPLE: 632.9", + "metrics.result": 551.221, + "metrics.result_per_accelerator": 68.902625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 3, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 3, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Gold 6438N", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 512.0, - "system.memory.configuration": "16x 32GB M321R4GA3BB6-CQKDS", - "system.name": "ThinkSystem SR650 V3 (3x NVIDIA L40S, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0516 ROUGE2: 20.0939 ROUGEL: 29.9288 GEN_LEN: 4111576", - "metrics.result": 2475.22, - "metrics.result_per_accelerator": 825.0733333333333, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.654998586911255 exact_match: 90.01192256783585 TOKENS_PER_SAMPLE: 646.4", + "metrics.result": 290.093, + "metrics.result_per_accelerator": 36.261625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 3, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 3, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Gold 6438N", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 512.0, - "system.memory.configuration": "16x 32GB M321R4GA3BB6-CQKDS", - "system.name": "ThinkSystem SR650 V3 (3x NVIDIA L40S, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0394 ROUGE2: 20.0943 ROUGEL: 29.9329 GEN_LEN: 4100159", - "metrics.result": 14073.6, - "metrics.result_per_accelerator": 1759.2, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1334 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 35192.5, + "metrics.result_per_accelerator": 4399.0625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 84, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9634 84-Core Processor", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB HMCT04MEERA131N", - "system.name": "ThinkSystem SR675 V3 (8x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0404 ROUGE2: 20.0949 ROUGEL: 29.9329 GEN_LEN: 4100496", - "metrics.result": 13083.5, - "metrics.result_per_accelerator": 1635.4375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5031 ROUGE2: 22.1063 ROUGEL: 28.7341 TOKENS_PER_SAMPLE: 292.5", + "metrics.result": 19413.2, + "metrics.result_per_accelerator": 2426.65, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 84, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9634 84-Core Processor", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB HMCT04MEERA131N", - "system.name": "ThinkSystem SR675 V3 (8x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5999 ROUGE2: 22.1861 ROUGEL: 28.805 TOKENS_PER_SAMPLE: 291.4", - "metrics.result": 15875.8, - "metrics.result_per_accelerator": 1984.475, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4192 ROUGE2: 23.1932 ROUGEL: 30.2622 TOKENS_PER_SAMPLE: 145.7 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.12", + "metrics.result": 62766.2, + "metrics.result_per_accelerator": 7845.775, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 84, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9634 84-Core Processor", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB HMCT04MEERA131N", - "system.name": "ThinkSystem SR675 V3 (8x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5887 ROUGE2: 22.1721 ROUGEL: 28.7905 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 13074.7, - "metrics.result_per_accelerator": 1634.3375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.518 ROUGE2: 23.1971 ROUGEL: 30.3919 TOKENS_PER_SAMPLE: 145.8 gsm8k_accuracy: 73.0 mbxp_accuracy: 59.7", + "metrics.result": 59647.1, + "metrics.result_per_accelerator": 7455.8875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 84, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9634 84-Core Processor", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB HMCT04MEERA131N", - "system.name": "ThinkSystem SR675 V3 (8x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5999 ROUGE2: 22.1861 ROUGEL: 28.805 TOKENS_PER_SAMPLE: 291.4", - "metrics.result": 15875.8, - "metrics.result_per_accelerator": 1984.475, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7836 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 35026.8, + "metrics.result_per_accelerator": 4378.35, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 84, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9634 84-Core Processor", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB HMCT04MEERA131N", - "system.name": "ThinkSystem SR675 V3 (8x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5887 ROUGE2: 22.1721 ROUGEL: 28.7905 TOKENS_PER_SAMPLE: 291.7", - "metrics.result": 13074.7, - "metrics.result_per_accelerator": 1634.3375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5599 ROUGE2: 22.1445 ROUGEL: 28.7952 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 33045.7, + "metrics.result_per_accelerator": 4130.7125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.5", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.5", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Lenovo", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 84, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9634 84-Core Processor", + "system.cpu.model": "AMD EPYC 9654", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 3072.0, - "system.memory.configuration": "24x 128GB HMCT04MEERA131N", - "system.name": "ThinkSystem SR675 V3 (8x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 4800MT/s", + "system.name": "AS-4125GS-TNHR2-LCC (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.3329 ROUGE2: 22.0197 ROUGEL: 28.598 TOKENS_PER_SAMPLE: 300.5", - "metrics.result": 948.198, - "metrics.result_per_accelerator": 118.52475, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", + "metrics.result": 98782.4, + "metrics.result_per_accelerator": 12347.8, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.5.2", + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.5.2", - "software.operating_system": "Ubuntu 22.04 (linux-5.15.0-94-generic-glibc2.35)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NeuralMagic", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d cache: 1.3 MiB (40 instances), L1i cache: 1.3 MiB (40 instances), L2 cache: 40 MiB (40 instances), L3 cache: 160 MiB (5 instances)", - "system.cpu.core_count": 4, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9254 24-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": null, - "system.memory.configuration": "undefined", - "system.name": "Crusoe Cloud L40S (8x L40S PCIe, vLLM, FP8)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4766 ROUGE2: 22.0865 ROUGEL: 28.6953 TOKENS_PER_SAMPLE: 295.9", - "metrics.result": 592.265, - "metrics.result_per_accelerator": 74.033125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7848 ROUGE2: 22.3844 ROUGEL: 29.1432 TOKENS_PER_SAMPLE: 273.8", + "metrics.result": 97629.8, + "metrics.result_per_accelerator": 12203.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.5.2", + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.5.2", - "software.operating_system": "Ubuntu 22.04 (linux-5.15.0-94-generic-glibc2.35)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NeuralMagic", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d cache: 1.3 MiB (40 instances), L1i cache: 1.3 MiB (40 instances), L2 cache: 40 MiB (40 instances), L3 cache: 160 MiB (5 instances)", - "system.cpu.core_count": 4, + "system.cpu.caches": null, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9254 24-Core Processor", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": null, - "system.memory.configuration": "undefined", - "system.name": "Crusoe Cloud L40S (8x L40S PCIe, vLLM, FP8)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-5", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7986 ROUGE2: 22.3524 ROUGEL: 29.161 TOKENS_PER_SAMPLE: 274.0", - "metrics.result": 11264.4, - "metrics.result_per_accelerator": 11264.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.8033 ROUGE2: 22.3646 ROUGEL: 29.1684 TOKENS_PER_SAMPLE: 271.8", + "metrics.result": 93857.0, + "metrics.result_per_accelerator": 11732.125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.1.0, CUDA 12.7", + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.1.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 22.04.5 LTS", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 1, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 12, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Silver 4410Y", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 408.0, - "system.memory.configuration": "6x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA B200 (1x B200-SXM-180GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.8023 ROUGE2: 22.355 ROUGEL: 29.1623 TOKENS_PER_SAMPLE: 273.9", - "metrics.result": 10755.6, - "metrics.result_per_accelerator": 10755.6, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.8002 ROUGE2: 22.3621 ROUGEL: 29.1653 TOKENS_PER_SAMPLE: 272.8", + "metrics.result": 59505.3, + "metrics.result_per_accelerator": 7438.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.1.0, CUDA 12.7", + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.1.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 22.04.5 LTS", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 1, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 12, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Silver 4410Y", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 408.0, - "system.memory.configuration": "6x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA B200 (1x B200-SXM-180GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.7986 ROUGE2: 22.3524 ROUGEL: 29.161 TOKENS_PER_SAMPLE: 274.0", - "metrics.result": 11264.4, - "metrics.result_per_accelerator": 11264.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 22.092192082812325 exact_match: 90.10742885506288 TOKENS_PER_SAMPLE: 654.1", + "metrics.result": 1538.17, + "metrics.result_per_accelerator": 192.27125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.1.0, CUDA 12.7", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.1.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 22.04.5 LTS", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 1, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 12, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Silver 4410Y", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 408.0, - "system.memory.configuration": "6x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA B200 (1x B200-SXM-180GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.8023 ROUGE2: 22.355 ROUGEL: 29.1623 TOKENS_PER_SAMPLE: 273.9", - "metrics.result": 10755.6, - "metrics.result_per_accelerator": 10755.6, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 22.044077554195237 exact_match: 90.11247518199869 TOKENS_PER_SAMPLE: 657.3", + "metrics.result": 1057.52, + "metrics.result_per_accelerator": 132.19, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.1.0, CUDA 12.7", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.7", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.1.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", - "submission.availability": "preview", + "software.operating_system": "Ubuntu 22.04.5 LTS", + "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 180.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA B200-SXM-180GB", - "system.accelerator.total_count": 1, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 12, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Silver 4410Y", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 408.0, - "system.memory.configuration": "6x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA B200 (1x B200-SXM-180GB, TensorRT)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 19739.8, - "metrics.result_per_accelerator": 2467.475, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.8033 ROUGE2: 22.3646 ROUGEL: 29.1684 TOKENS_PER_SAMPLE: 271.8", + "metrics.result": 93857.0, + "metrics.result_per_accelerator": 11732.125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0996 ROUGE2: 20.1107 ROUGEL: 29.9733 GEN_LEN: 4114082", - "metrics.result": 19233.8, - "metrics.result_per_accelerator": 2404.225, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.8002 ROUGE2: 22.3621 ROUGEL: 29.1653 TOKENS_PER_SAMPLE: 272.8", + "metrics.result": 59505.3, + "metrics.result_per_accelerator": 7438.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 19739.8, - "metrics.result_per_accelerator": 2467.475, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.6325 ROUGE2: 23.3874 ROUGEL: 30.5099 TOKENS_PER_SAMPLE: 144.4 gsm8k_accuracy: 73.3 mbxp_accuracy: 60.18", + "metrics.result": 128795.0, + "metrics.result_per_accelerator": 16099.375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0996 ROUGE2: 20.1107 ROUGEL: 29.9733 GEN_LEN: 4114082", - "metrics.result": 19233.8, - "metrics.result_per_accelerator": 2404.225, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4541 ROUGE2: 23.1973 ROUGEL: 30.2838 TOKENS_PER_SAMPLE: 146.3 gsm8k_accuracy: 72.94 mbxp_accuracy: 59.96", + "metrics.result": 129047.0, + "metrics.result_per_accelerator": 16130.875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5437 ROUGE2: 22.1222 ROUGEL: 28.7396 TOKENS_PER_SAMPLE: 290.3", - "metrics.result": 24524.9, - "metrics.result_per_accelerator": 3065.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7869 ROUGE2: 22.383 ROUGEL: 29.1434 TOKENS_PER_SAMPLE: 272.9", + "metrics.result": 98782.4, + "metrics.result_per_accelerator": 12347.8, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, - { - "metrics.accuracy": "ROUGE1: 44.5435 ROUGE2: 22.122 ROUGEL: 28.7395 TOKENS_PER_SAMPLE: 290.3", - "metrics.result": 21605.8, - "metrics.result_per_accelerator": 2700.725, + { + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.7848 ROUGE2: 22.3844 ROUGEL: 29.1432 TOKENS_PER_SAMPLE: 273.8", + "metrics.result": 97629.8, + "metrics.result_per_accelerator": 12203.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp4", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 22.04.5 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 180.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA B200-SXM-180GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8562Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink-5", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5 5600MT/s @4400MT/s", + "system.name": "SYS-421GE-NBRT-LCC (8x B200-SXM-180GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5437 ROUGE2: 22.1222 ROUGEL: 28.7396 TOKENS_PER_SAMPLE: 290.3", - "metrics.result": 24524.9, - "metrics.result_per_accelerator": 3065.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 18724.2, + "metrics.result_per_accelerator": 2340.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) 6979P", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "SYS-522GA-NRT(8x H200-NVL-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5435 ROUGE2: 22.122 ROUGEL: 28.7395 TOKENS_PER_SAMPLE: 290.3", - "metrics.result": 21605.8, - "metrics.result_per_accelerator": 2700.725, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 18571.1, + "metrics.result_per_accelerator": 2321.3875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) 6979P", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "SYS-522GA-NRT(8x H200-NVL-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3448 ROUGE2: 23.2415 ROUGEL: 30.3087 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.12 mbxp_accuracy: 59.88", - "metrics.result": 52415.9, - "metrics.result_per_accelerator": 6551.9875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 18724.2, + "metrics.result_per_accelerator": 2340.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) 6979P", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "SYS-522GA-NRT(8x H200-NVL-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3537 ROUGE2: 23.2446 ROUGEL: 30.3114 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", - "metrics.result": 50795.6, - "metrics.result_per_accelerator": 6349.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 18571.1, + "metrics.result_per_accelerator": 2321.3875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) 6979P", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA DGX H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "SYS-522GA-NRT(8x H200-NVL-141GB)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 2627.69, - "metrics.result_per_accelerator": 2627.69, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.3988 ROUGE2: 22.039 ROUGEL: 28.6453 TOKENS_PER_SAMPLE: 295.3", + "metrics.result": 33069.1, + "metrics.result_per_accelerator": 4133.6375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": null, + "software.framework": "vLLM 0.0.0.dev1+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.7.0", + "software.framework.rocm": "6.3.1", "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.0.0", + "software.operating_system": "Ubuntu 22.04 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_config": "HBM3E", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 64, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 6400MT/s @6000MT/s", + "system.name": "AS-8126GS-TNMR", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1002 ROUGE2: 20.1114 ROUGEL: 29.9742 GEN_LEN: 4114259", - "metrics.result": 2513.38, - "metrics.result_per_accelerator": 2513.38, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.46 ROUGE2: 22.0494 ROUGEL: 28.6787 TOKENS_PER_SAMPLE: 295.2", + "metrics.result": 28770.2, + "metrics.result_per_accelerator": 3596.275, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": null, + "software.framework": "vLLM 0.0.0.dev1+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.7.0", + "software.framework.rocm": "6.3.1", "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.0.0", + "software.operating_system": "Ubuntu 22.04 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_config": "HBM3E", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 64, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 6400MT/s @6000MT/s", + "system.name": "AS-8126GS-TNMR", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 2627.69, - "metrics.result_per_accelerator": 2627.69, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.3988 ROUGE2: 22.039 ROUGEL: 28.6453 TOKENS_PER_SAMPLE: 295.3", + "metrics.result": 33069.1, + "metrics.result_per_accelerator": 4133.6375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": null, + "software.framework": "vLLM 0.0.0.dev1+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.7.0", + "software.framework.rocm": "6.3.1", "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.0.0", + "software.operating_system": "Ubuntu 22.04 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_config": "HBM3E", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 64, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 6400MT/s @6000MT/s", + "system.name": "AS-8126GS-TNMR", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1002 ROUGE2: 20.1114 ROUGEL: 29.9742 GEN_LEN: 4114259", - "metrics.result": 2513.38, - "metrics.result_per_accelerator": 2513.38, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.46 ROUGE2: 22.0494 ROUGEL: 28.6787 TOKENS_PER_SAMPLE: 295.2", + "metrics.result": 28770.2, + "metrics.result_per_accelerator": 3596.275, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": null, + "software.framework": "vLLM 0.0.0.dev1+mlperf50, Pytorch 2.7.0a0+git3a58512, ROCm 6.3.1", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": null, "software.framework.cudnn": null, "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, + "software.framework.pytorch": "2.7.0", + "software.framework.rocm": "6.3.1", "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.framework.tensorrt": null, + "software.framework.vllm": "0.0.0", + "software.operating_system": "Ubuntu 22.04 LTS", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Supermicro", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, - "system.accelerator.vendor": "NVIDIA", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 256.0, + "system.accelerator.memory_config": "HBM3E", + "system.accelerator.name": "AMD Instinct MI325X 256GB HBM3E", + "system.accelerator.total_count": 8, + "system.accelerator.vendor": "AMD", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 64, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "AMD EPYC 9575F", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "XGMI", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2355.2, + "system.memory.configuration": "24x 96GB DDR5 6400MT/s @6000MT/s", + "system.name": "AS-8126GS-TNMR", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 4067.52, - "metrics.result_per_accelerator": 4067.52, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6821 ROUGE2: 22.2379 ROUGEL: 28.8722 TOKENS_PER_SAMPLE: 289.4", + "metrics.result": 18084.9, + "metrics.result_per_accelerator": 2260.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1576 ROUGEL: 28.7742 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 3883.67, - "metrics.result_per_accelerator": 3883.67, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6735 ROUGE2: 22.2241 ROUGEL: 28.8629 TOKENS_PER_SAMPLE: 290.3", + "metrics.result": 16848.6, + "metrics.result_per_accelerator": 2106.075, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 4067.52, - "metrics.result_per_accelerator": 4067.52, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 14769.7, + "metrics.result_per_accelerator": 1846.2125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1576 ROUGEL: 28.7742 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 3883.67, - "metrics.result_per_accelerator": 3883.67, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 14846.7, + "metrics.result_per_accelerator": 1855.8375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3572 ROUGE2: 23.2487 ROUGEL: 30.3161 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.92", - "metrics.result": 8021.46, - "metrics.result_per_accelerator": 8021.46, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 14769.7, + "metrics.result_per_accelerator": 1846.2125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3563 ROUGE2: 23.2486 ROUGEL: 30.3161 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.94", - "metrics.result": 7450.27, - "metrics.result_per_accelerator": 7450.27, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 14846.7, + "metrics.result_per_accelerator": 1855.8375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 144.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 144GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200 NVL2 Platform (1x GH200-144GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 4487.88, - "metrics.result_per_accelerator": 4487.88, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6743 ROUGE2: 22.2269 ROUGEL: 28.8649 TOKENS_PER_SAMPLE: 289.3", + "metrics.result": 15553.3, + "metrics.result_per_accelerator": 1944.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", - "system.name": "NVIDIA H200 (1x H200-SXM-141GB-CTS, TensorRT)", + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5629 ROUGE2: 22.1577 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 4202.3, - "metrics.result_per_accelerator": 4202.3, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6131 ROUGE2: 22.198 ROUGEL: 28.8436 TOKENS_PER_SAMPLE: 290.5", + "metrics.result": 11041.5, + "metrics.result_per_accelerator": 1380.1875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", - "system.name": "NVIDIA H200 (1x H200-SXM-141GB-CTS, TensorRT)", + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 4487.88, - "metrics.result_per_accelerator": 4487.88, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6743 ROUGE2: 22.2269 ROUGEL: 28.8649 TOKENS_PER_SAMPLE: 289.3", + "metrics.result": 15553.3, + "metrics.result_per_accelerator": 1944.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", - "system.name": "NVIDIA H200 (1x H200-SXM-141GB-CTS, TensorRT)", + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5629 ROUGE2: 22.1577 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 4202.3, - "metrics.result_per_accelerator": 4202.3, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6131 ROUGE2: 22.198 ROUGEL: 28.8436 TOKENS_PER_SAMPLE: 290.5", + "metrics.result": 11041.5, + "metrics.result_per_accelerator": 1380.1875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", - "system.name": "NVIDIA H200 (1x H200-SXM-141GB-CTS, TensorRT)", + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 34864.2, - "metrics.result_per_accelerator": 4358.025, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.451 ROUGE2: 23.2629 ROUGEL: 30.3646 TOKENS_PER_SAMPLE: 144.9 gsm8k_accuracy: 73.42 mbxp_accuracy: 59.9", + "metrics.result": 37713.6, + "metrics.result_per_accelerator": 4714.2, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB-CTS, TensorRT)", + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5618 ROUGE2: 22.1563 ROUGEL: 28.7731 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 32789.7, - "metrics.result_per_accelerator": 4098.7125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4195 ROUGE2: 23.275 ROUGEL: 30.3242 TOKENS_PER_SAMPLE: 146.4 gsm8k_accuracy: 73.04 mbxp_accuracy: 59.94", + "metrics.result": 37934.5, + "metrics.result_per_accelerator": 4741.8125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB-CTS, TensorRT)", + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5631 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 34864.2, - "metrics.result_per_accelerator": 4358.025, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6821 ROUGE2: 22.2379 ROUGEL: 28.8722 TOKENS_PER_SAMPLE: 289.4", + "metrics.result": 18084.9, + "metrics.result_per_accelerator": 2260.6125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB-CTS, TensorRT)", + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5618 ROUGE2: 22.1563 ROUGEL: 28.7731 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 32789.7, - "metrics.result_per_accelerator": 4098.7125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.6735 ROUGE2: 22.2241 ROUGEL: 28.8629 TOKENS_PER_SAMPLE: 290.3", + "metrics.result": 16848.6, + "metrics.result_per_accelerator": 2106.075, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Fujitsu", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB-CTS", + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 3MiB per core; L1i: 2MiB per core; L2: 128MiB per core; L3: 320MiB per chip", + "system.cpu.core_count": 32, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.frequency": 4.0, + "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA12", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB-CTS, TensorRT)", + "system.memory.configuration": "16x 64GB M321R8GA0BB0-CQKZJ (64GB 4800MHz)", + "system.name": "PRIMERGY CDI (8x H100NVL, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 2579.51, - "metrics.result_per_accelerator": 2579.51, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5269 ROUGE2: 22.1161 ROUGEL: 28.7612 TOKENS_PER_SAMPLE: 290.6", + "metrics.result": 17654.1, + "metrics.result_per_accelerator": 4413.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", + "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 1, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "AMD EPYC 9655 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (1x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB M321R8GA0PB2", + "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1016 ROUGE2: 20.1124 ROUGEL: 29.9751 GEN_LEN: 4114376", - "metrics.result": 2405.86, - "metrics.result_per_accelerator": 2405.86, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5272 ROUGE2: 22.1154 ROUGEL: 28.7622 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 14652.8, + "metrics.result_per_accelerator": 3663.2, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", + "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 1, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "AMD EPYC 9655 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (1x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB M321R8GA0PB2", + "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 2579.51, - "metrics.result_per_accelerator": 2579.51, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 10825.5, + "metrics.result_per_accelerator": 2706.375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", + "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 1, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "AMD EPYC 9655 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (1x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB M321R8GA0PB2", + "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1016 ROUGE2: 20.1124 ROUGEL: 29.9751 GEN_LEN: 4114376", - "metrics.result": 2405.86, - "metrics.result_per_accelerator": 2405.86, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 10503.7, + "metrics.result_per_accelerator": 2625.925, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99.9", "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", + "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 1, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "AMD EPYC 9655 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (1x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB M321R8GA0PB2", + "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 20086.1, - "metrics.result_per_accelerator": 2510.7625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 10825.5, + "metrics.result_per_accelerator": 2706.375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "AMD EPYC 9655 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB M321R8GA0PB2", + "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1027 ROUGE2: 20.1139 ROUGEL: 29.9764 GEN_LEN: 4113784", - "metrics.result": 19243.3, - "metrics.result_per_accelerator": 2405.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 10503.7, + "metrics.result_per_accelerator": 2625.925, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "AMD EPYC 9655 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB M321R8GA0PB2", + "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 20086.1, - "metrics.result_per_accelerator": 2510.7625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4435 ROUGE2: 23.2094 ROUGEL: 30.2857 TOKENS_PER_SAMPLE: 145.3 gsm8k_accuracy: 73.3 mbxp_accuracy: 59.86", + "metrics.result": 31503.7, + "metrics.result_per_accelerator": 7875.925, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "AMD EPYC 9655 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB M321R8GA0PB2", + "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1027 ROUGE2: 20.1139 ROUGEL: 29.9764 GEN_LEN: 4113784", - "metrics.result": 19243.3, - "metrics.result_per_accelerator": 2405.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.416 ROUGE2: 23.2389 ROUGEL: 30.246 TOKENS_PER_SAMPLE: 146.8 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", + "metrics.result": 29301.5, + "metrics.result_per_accelerator": 7325.375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "AMD EPYC 9655 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB M321R8GA0PB2", + "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 31302.7, - "metrics.result_per_accelerator": 3912.8375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5269 ROUGE2: 22.1161 ROUGEL: 28.7612 TOKENS_PER_SAMPLE: 290.6", + "metrics.result": 17654.1, + "metrics.result_per_accelerator": 4413.525, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "AMD EPYC 9655 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB M321R8GA0PB2", + "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5616 ROUGE2: 22.1566 ROUGEL: 28.7725 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 29228.2, - "metrics.result_per_accelerator": 3653.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5272 ROUGE2: 22.1154 ROUGEL: 28.7622 TOKENS_PER_SAMPLE: 291.7", + "metrics.result": 14652.8, + "metrics.result_per_accelerator": 3663.2, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, + "system.accelerator.count_per_node": 4, "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 96, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "AMD EPYC 9655 96-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", + "system.cpu.vendor": "AMD", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.configuration": "32x 64GB M321R8GA0PB2", + "system.name": "ThinkSystem SR675 V3 (4x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1578 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 31302.7, - "metrics.result_per_accelerator": 3912.8375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 7705.97, + "metrics.result_per_accelerator": 1926.4925, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 8.1 MiB, L1i: 10.8 MiB, L2: 344 MiB, L3: 672 MiB", + "system.cpu.core_count": 86, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) 6787P", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB HMCG94BDJHA464N", + "system.name": "ThinkSystem SR650a V4 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5616 ROUGE2: 22.1566 ROUGEL: 28.7725 TOKENS_PER_SAMPLE: 291.3", - "metrics.result": 29228.2, - "metrics.result_per_accelerator": 3653.525, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 6773.87, + "metrics.result_per_accelerator": 1693.4675, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 8.1 MiB, L1i: 10.8 MiB, L2: 344 MiB, L3: 672 MiB", + "system.cpu.core_count": 86, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) 6787P", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB HMCG94BDJHA464N", + "system.name": "ThinkSystem SR650a V4 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3568 ROUGE2: 23.2496 ROUGEL: 30.3171 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", - "metrics.result": 59022.3, - "metrics.result_per_accelerator": 7377.7875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 7705.97, + "metrics.result_per_accelerator": 1926.4925, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 8.1 MiB, L1i: 10.8 MiB, L2: 344 MiB, L3: 672 MiB", + "system.cpu.core_count": 86, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) 6787P", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB HMCG94BDJHA464N", + "system.name": "ThinkSystem SR650a V4 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3535 ROUGE2: 23.2468 ROUGEL: 30.3146 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.12 mbxp_accuracy: 59.9", - "metrics.result": 57177.2, - "metrics.result_per_accelerator": 7147.15, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 6773.87, + "metrics.result_per_accelerator": 1693.4675, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 8.1 MiB, L1i: 10.8 MiB, L2: 344 MiB, L3: 672 MiB", + "system.cpu.core_count": 86, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) 6787P", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB HMCG94BDJHA464N", + "system.name": "ThinkSystem SR650a V4 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 13096.6, - "metrics.result_per_accelerator": 1637.075, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.3468 ROUGE2: 23.202 ROUGEL: 30.3039 TOKENS_PER_SAMPLE: 145.4 gsm8k_accuracy: 73.34 mbxp_accuracy: 60.02", + "metrics.result": 19500.9, + "metrics.result_per_accelerator": 4875.225, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 8.1 MiB, L1i: 10.8 MiB, L2: 344 MiB, L3: 672 MiB", + "system.cpu.core_count": 86, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) 6787P", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB HMCG94BDJHA464N", + "system.name": "ThinkSystem SR650a V4 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0987 ROUGE2: 20.1115 ROUGEL: 29.9736 GEN_LEN: 4114209", - "metrics.result": 11700.8, - "metrics.result_per_accelerator": 1462.6, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4582 ROUGE2: 23.1907 ROUGEL: 30.36 TOKENS_PER_SAMPLE: 146.9 gsm8k_accuracy: 72.98 mbxp_accuracy: 59.74", + "metrics.result": 19334.9, + "metrics.result_per_accelerator": 4833.725, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 8, + "system.accelerator.count_per_node": 4, + "system.accelerator.memory_capacity": 94.0, + "system.accelerator.memory_config": "HBM3", + "system.accelerator.name": "NVIDIA H100-NVL-94GB", + "system.accelerator.total_count": 4, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.caches": "L1d: 8.1 MiB, L1i: 10.8 MiB, L2: 344 MiB, L3: 672 MiB", + "system.cpu.core_count": 86, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "Intel(R) Xeon(R) 6787P", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "16x 64GB HMCG94BDJHA464N", + "system.name": "ThinkSystem SR650a V4 (4x H100-NVL-94GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 13096.6, - "metrics.result_per_accelerator": 1637.075, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7628 TOKENS_PER_SAMPLE: 290.7", + "metrics.result": 35400.2, + "metrics.result_per_accelerator": 4425.025, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -34623,44 +37029,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0987 ROUGE2: 20.1115 ROUGEL: 29.9736 GEN_LEN: 4114209", - "metrics.result": 11700.8, - "metrics.result_per_accelerator": 1462.6, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5214 ROUGE2: 22.1163 ROUGEL: 28.7576 TOKENS_PER_SAMPLE: 291.6", + "metrics.result": 32934.7, + "metrics.result_per_accelerator": 4116.8375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -34669,44 +37078,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1578 ROUGEL: 28.7741 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 25262.1, - "metrics.result_per_accelerator": 3157.7625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 21626.0, + "metrics.result_per_accelerator": 2703.25, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -34715,44 +37127,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5622 ROUGE2: 22.1565 ROUGEL: 28.7734 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 23113.1, - "metrics.result_per_accelerator": 2889.1375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 21553.4, + "metrics.result_per_accelerator": 2694.175, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -34761,44 +37176,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.563 ROUGE2: 22.1578 ROUGEL: 28.7741 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 25262.1, - "metrics.result_per_accelerator": 3157.7625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 21626.0, + "metrics.result_per_accelerator": 2703.25, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -34807,44 +37225,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5622 ROUGE2: 22.1565 ROUGEL: 28.7734 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 23113.1, - "metrics.result_per_accelerator": 2889.1375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 21553.4, + "metrics.result_per_accelerator": 2694.175, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -34853,44 +37274,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3568 ROUGE2: 23.2496 ROUGEL: 30.3171 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", - "metrics.result": 48987.9, - "metrics.result_per_accelerator": 6123.4875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7629 TOKENS_PER_SAMPLE: 290.7", + "metrics.result": 35498.5, + "metrics.result_per_accelerator": 4437.3125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -34899,44 +37323,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3548 ROUGE2: 23.2479 ROUGEL: 30.3149 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.12 mbxp_accuracy: 59.88", - "metrics.result": 45497.0, - "metrics.result_per_accelerator": 5687.125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.576 ROUGE2: 22.1459 ROUGEL: 28.7741 TOKENS_PER_SAMPLE: 292.6", + "metrics.result": 19448.1, + "metrics.result_per_accelerator": 2431.0125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -34945,44 +37372,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, MaxQ, TensorRT)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5638 ROUGE2: 22.1581 ROUGEL: 28.7745 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 31059.3, - "metrics.result_per_accelerator": 3882.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7629 TOKENS_PER_SAMPLE: 290.7", + "metrics.result": 35498.5, + "metrics.result_per_accelerator": 4437.3125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -34991,44 +37421,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5623 ROUGE2: 22.1574 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 292.2", - "metrics.result": 30128.4, - "metrics.result_per_accelerator": 3766.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.576 ROUGE2: 22.1459 ROUGEL: 28.7741 TOKENS_PER_SAMPLE: 292.6", + "metrics.result": 19448.1, + "metrics.result_per_accelerator": 2431.0125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -35037,44 +37470,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5638 ROUGE2: 22.1581 ROUGEL: 28.7745 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 31059.3, - "metrics.result_per_accelerator": 3882.4125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4086 ROUGE2: 23.1851 ROUGEL: 30.262 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.4 mbxp_accuracy: 60.12", + "metrics.result": 63514.7, + "metrics.result_per_accelerator": 7939.3375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -35083,44 +37519,47 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5623 ROUGE2: 22.1574 ROUGEL: 28.774 TOKENS_PER_SAMPLE: 292.2", - "metrics.result": 30128.4, - "metrics.result_per_accelerator": 3766.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4078 ROUGE2: 23.1492 ROUGEL: 30.338 TOKENS_PER_SAMPLE: 144.7 gsm8k_accuracy: 73.02 mbxp_accuracy: 60.0", + "metrics.result": 61074.2, + "metrics.result_per_accelerator": 7634.275, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "NVIDIA", + "submission.organization": "Lenovo", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 141.0, @@ -35129,813 +37568,818 @@ "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (8x H200-SXM-141GB, TensorRT, Triton)", + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 42.9198 ROUGE2: 20.1178 ROUGEL: 29.9386 GEN_LEN: 4050052", - "metrics.result": 64.4734, - "metrics.result_per_accelerator": 64.4734, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "Jetpack 6.0, TensorRT 10.1, CUDA 12.2", - "software.framework.ai": null, - "software.framework.cuda": "12.2", - "software.framework.cudnn": null, - "software.framework.jetpack": "6.0", - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.1", - "software.framework.vllm": null, - "software.operating_system": "Jetson r36.3.1 L4T", - "submission.availability": "available", - "submission.division": "closed", - "submission.organization": "NVIDIA", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": null, - "system.accelerator.memory_config": "LPDDR5", - "system.accelerator.name": "NVIDIA Jetson AGX Orin 64G", - "system.accelerator.total_count": 1, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": null, - "system.cpu.core_count": 12, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": null, - "system.cpu.model": "12-core ARM Cortex-A78AE CPU", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "ARM", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 64.0, - "system.memory.configuration": "64GB 256-bit LPDDR5", - "system.name": "NVIDIA Jetson AGX Orin Developer Kit 64G (TensorRT)", - "system.number_of_nodes": 1, - "system.type": "edge" - }, - { - "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", - "metrics.result": 2695.15, - "metrics.result_per_accelerator": 2695.15, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7628 TOKENS_PER_SAMPLE: 290.7", + "metrics.result": 35400.2, + "metrics.result_per_accelerator": 4425.025, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 512.0, - "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", - "metrics.result": 2159.58, - "metrics.result_per_accelerator": 2159.58, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5214 ROUGE2: 22.1163 ROUGEL: 28.7576 TOKENS_PER_SAMPLE: 291.6", + "metrics.result": 32934.7, + "metrics.result_per_accelerator": 4116.8375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 512.0, - "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB HMCG94AGBRA181N", + "system.name": "ThinkSystem SR780a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", - "metrics.result": 2695.15, - "metrics.result_per_accelerator": 2695.15, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7628 TOKENS_PER_SAMPLE: 290.7", + "metrics.result": 35453.0, + "metrics.result_per_accelerator": 4431.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 512.0, - "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", - "metrics.result": 2159.58, - "metrics.result_per_accelerator": 2159.58, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5225 ROUGE2: 22.1121 ROUGEL: 28.7551 TOKENS_PER_SAMPLE: 291.6", + "metrics.result": 32976.4, + "metrics.result_per_accelerator": 4122.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Oracle", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 512.0, - "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0913 ROUGE2: 20.1321 ROUGEL: 29.9767 GEN_LEN: 4113606", - "metrics.result": 3463.19, - "metrics.result_per_accelerator": 865.7975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 21513.3, + "metrics.result_per_accelerator": 2689.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0.19", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 44, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8458P", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 16", - "system.name": "D54U_3U_L40S_PCIe_48GBx4_TRT", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0892 ROUGE2: 20.1306 ROUGEL: 29.9753 GEN_LEN: 4113194", - "metrics.result": 3096.45, - "metrics.result_per_accelerator": 774.1125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 21549.1, + "metrics.result_per_accelerator": 2693.6375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0.19", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 44, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8458P", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 16", - "system.name": "D54U_3U_L40S_PCIe_48GBx4_TRT", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0913 ROUGE2: 20.1321 ROUGEL: 29.9767 GEN_LEN: 4113606", - "metrics.result": 3463.19, - "metrics.result_per_accelerator": 865.7975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 21513.3, + "metrics.result_per_accelerator": 2689.1625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0.19", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 44, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8458P", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 16", - "system.name": "D54U_3U_L40S_PCIe_48GBx4_TRT", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0892 ROUGE2: 20.1306 ROUGEL: 29.9753 GEN_LEN: 4113194", - "metrics.result": 3096.45, - "metrics.result_per_accelerator": 774.1125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0858 ROUGE2: 20.1151 ROUGEL: 29.9685 GEN_LEN: 4153033", + "metrics.result": 21549.1, + "metrics.result_per_accelerator": 2693.6375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0.19", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "GDDR6", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 44, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8458P", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe", - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": 1024.0, - "system.memory.configuration": " DDR4-4800 64GB x 16", - "system.name": "D54U_3U_L40S_PCIe_48GBx4_TRT", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0658 ROUGE2: 20.1299 ROUGEL: 29.9759 GEN_LEN: 4108404", - "metrics.result": 2803.72, - "metrics.result_per_accelerator": 2803.72, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7628 TOKENS_PER_SAMPLE: 290.7", + "metrics.result": 35383.8, + "metrics.result_per_accelerator": 4422.975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0.19", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 512.0, - "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", - "system.name": "GH200-GraceHopper-Superchip_GH200-96GB_aarch64x1_TRT", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0658 ROUGE2: 20.1298 ROUGEL: 29.9758 GEN_LEN: 4108405", - "metrics.result": 2160.12, - "metrics.result_per_accelerator": 2160.12, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5762 ROUGE2: 22.1484 ROUGEL: 28.7736 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 19492.5, + "metrics.result_per_accelerator": 2436.5625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0.19", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 512.0, - "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", - "system.name": "GH200-GraceHopper-Superchip_GH200-96GB_aarch64x1_TRT", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0658 ROUGE2: 20.1299 ROUGEL: 29.9759 GEN_LEN: 4108404", - "metrics.result": 2803.72, - "metrics.result_per_accelerator": 2803.72, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7628 TOKENS_PER_SAMPLE: 290.7", + "metrics.result": 35383.8, + "metrics.result_per_accelerator": 4422.975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0.19", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 512.0, - "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", - "system.name": "GH200-GraceHopper-Superchip_GH200-96GB_aarch64x1_TRT", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0658 ROUGE2: 20.1298 ROUGEL: 29.9758 GEN_LEN: 4108405", - "metrics.result": 2160.12, - "metrics.result_per_accelerator": 2160.12, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5762 ROUGE2: 22.1484 ROUGEL: 28.7736 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 19492.5, + "metrics.result_per_accelerator": 2436.5625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0.19", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 512.0, - "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", - "system.name": "GH200-GraceHopper-Superchip_GH200-96GB_aarch64x1_TRT", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5991 ROUGE2: 22.1726 ROUGEL: 28.8022 TOKENS_PER_SAMPLE: 291.8", - "metrics.result": 3114.03, - "metrics.result_per_accelerator": 3114.03, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4074 ROUGE2: 23.2168 ROUGEL: 30.2703 TOKENS_PER_SAMPLE: 145.3 gsm8k_accuracy: 73.62 mbxp_accuracy: 60.24", + "metrics.result": 63446.7, + "metrics.result_per_accelerator": 7930.8375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0.19", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 512.0, - "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", - "system.name": "GH200-GraceHopper-Superchip_GH200-96GB_aarch64x1_TRT", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5947 ROUGE2: 22.1631 ROUGEL: 28.793 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 2619.09, - "metrics.result_per_accelerator": 2619.09, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.3802 ROUGE2: 23.1803 ROUGEL: 30.3072 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.0 mbxp_accuracy: 60.2", + "metrics.result": 61317.2, + "metrics.result_per_accelerator": 7664.65, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0.19, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0.19", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Rocky Linux 9.3", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Quanta_Cloud_Technology", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 512.0, - "system.memory.configuration": "16x 16DP (32GB) LPDDR5x", - "system.name": "GH200-GraceHopper-Superchip_GH200-96GB_aarch64x1_TRT", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6053 ROUGE2: 22.2332 ROUGEL: 28.8873 TOKENS_PER_SAMPLE: 291.4", - "metrics.result": 1717.77, - "metrics.result_per_accelerator": 429.4425, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5325 ROUGE2: 22.1163 ROUGEL: 28.7628 TOKENS_PER_SAMPLE: 290.7", + "metrics.result": 35453.0, + "metrics.result_per_accelerator": 4431.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Red Hat Enterprise Linux CoreOS release 4.14", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "RedHat", + "submission.organization": "Lenovo", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "HBM2", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 112, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480CL", - "system.cpu.vcpu_count": "-", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB Micron DDR5", - "system.name": "L40S-RedHat-OpenShift", + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.4633 ROUGE2: 22.0929 ROUGEL: 28.7123 TOKENS_PER_SAMPLE: 295.3", - "metrics.result": 1469.19, - "metrics.result_per_accelerator": 367.2975, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5225 ROUGE2: 22.1121 ROUGEL: 28.7551 TOKENS_PER_SAMPLE: 291.6", + "metrics.result": 32976.4, + "metrics.result_per_accelerator": 4122.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Red Hat Enterprise Linux CoreOS release 4.14", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "RedHat", + "submission.organization": "Lenovo", "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 48.0, - "system.accelerator.memory_config": "HBM2", - "system.accelerator.name": "NVIDIA L40S", - "system.accelerator.total_count": 4, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 112, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480CL", - "system.cpu.vcpu_count": "-", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", + "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "PCIe", + "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB Micron DDR5", - "system.name": "L40S-RedHat-OpenShift", + "system.memory.configuration": "32x 64GB M321R8GA0PB0", + "system.name": "ThinkSystem SR680a V3(8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 19539.2, - "metrics.result_per_accelerator": 2442.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.3218 ROUGE2: 23.243 ROUGEL: 30.2997 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.08 mbxp_accuracy: 59.9", + "metrics.result": 51514.5, + "metrics.result_per_accelerator": 6439.3125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.4", "software.framework.ai": null, "software.framework.cuda": "12.4", "software.framework.cudnn": null, @@ -35943,45 +38387,48 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04.1, VMware ESXi 8.0.3 2441450", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.name": "Virtualized NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 2.1, + "system.cpu.model": "Intel Xeon Platinum 8468", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MHz", - "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "TBD", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB,VMware ESXi 8.0.3) ", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1036 ROUGE2: 20.1145 ROUGEL: 29.9777 GEN_LEN: 4113966", - "metrics.result": 19810.7, - "metrics.result_per_accelerator": 2476.3375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.5359 ROUGE2: 23.2614 ROUGEL: 30.4398 TOKENS_PER_SAMPLE: 146.4 gsm8k_accuracy: 73.12 mbxp_accuracy: 59.92", + "metrics.result": 50896.0, + "metrics.result_per_accelerator": 6362.0, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.4", "software.framework.ai": null, "software.framework.cuda": "12.4", "software.framework.cudnn": null, @@ -35989,1898 +38436,2021 @@ "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04.1, VMware ESXi 8.0.3 2441450", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.name": "Virtualized NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.frequency": 2.1, + "system.cpu.model": "Intel Xeon Platinum 8468", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": null, "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MHz", - "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 1024.0, + "system.memory.configuration": "TBD", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB,VMware ESXi 8.0.3) ", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 19539.2, - "metrics.result_per_accelerator": 2442.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5457 ROUGE2: 22.133 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34763.2, + "metrics.result_per_accelerator": 4345.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MHz", - "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1036 ROUGE2: 20.1145 ROUGEL: 29.9777 GEN_LEN: 4113966", - "metrics.result": 19810.7, - "metrics.result_per_accelerator": 2476.3375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5602 ROUGE2: 22.142 ROUGEL: 28.794 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 33023.5, + "metrics.result_per_accelerator": 4127.9375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MHz", - "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.179 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 24216.8, - "metrics.result_per_accelerator": 3027.1, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21406.8, + "metrics.result_per_accelerator": 2675.85, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MHz", - "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1792 ROUGEL: 28.8001 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 23699.7, - "metrics.result_per_accelerator": 2962.4625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21455.9, + "metrics.result_per_accelerator": 2681.9875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MHz", - "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.179 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 24216.8, - "metrics.result_per_accelerator": 3027.1, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21406.8, + "metrics.result_per_accelerator": 2675.85, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MHz", - "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1792 ROUGEL: 28.8001 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 23699.7, - "metrics.result_per_accelerator": 2962.4625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21455.9, + "metrics.result_per_accelerator": 2681.9875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MHz", - "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3541 ROUGE2: 23.2468 ROUGEL: 30.3101 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.16 mbxp_accuracy: 59.88", - "metrics.result": 51269.0, - "metrics.result_per_accelerator": 6408.625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34538.4, + "metrics.result_per_accelerator": 4317.3, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MHz", - "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.351 ROUGE2: 23.2445 ROUGEL: 30.3143 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", - "metrics.result": 50099.5, - "metrics.result_per_accelerator": 6262.4375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5133 ROUGE2: 22.1131 ROUGEL: 28.7404 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 19415.9, + "metrics.result_per_accelerator": 2426.9875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 96, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9654", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2355.2, - "system.memory.configuration": "24x 96GB DDR5 4800MHz", - "system.name": "AS-4125GS-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 19418.9, - "metrics.result_per_accelerator": 2427.3625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.618544328060732 exact_match: 89.99261085373924 TOKENS_PER_SAMPLE: 646.3", + "metrics.result": 559.205, + "metrics.result_per_accelerator": 69.900625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9474F", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB DDR5 4800MHz", - "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0997 ROUGE2: 20.1116 ROUGEL: 29.9738 GEN_LEN: 4114213", - "metrics.result": 19304.4, - "metrics.result_per_accelerator": 2413.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.67381013091784 exact_match: 89.96315023163469 TOKENS_PER_SAMPLE: 650.6", + "metrics.result": 291.486, + "metrics.result_per_accelerator": 36.43575, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9474F", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB DDR5 4800MHz", - "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 19418.9, - "metrics.result_per_accelerator": 2427.3625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34538.4, + "metrics.result_per_accelerator": 4317.3, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9474F", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB DDR5 4800MHz", - "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0997 ROUGE2: 20.1116 ROUGEL: 29.9738 GEN_LEN: 4114213", - "metrics.result": 19304.4, - "metrics.result_per_accelerator": 2413.05, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5133 ROUGE2: 22.1131 ROUGEL: 28.7404 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 19415.9, + "metrics.result_per_accelerator": 2426.9875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9474F", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB DDR5 4800MHz", - "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 24011.7, - "metrics.result_per_accelerator": 3001.4625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.37 ROUGE2: 23.2496 ROUGEL: 30.2632 TOKENS_PER_SAMPLE: 145.6 gsm8k_accuracy: 73.3 mbxp_accuracy: 60.1", + "metrics.result": 62042.2, + "metrics.result_per_accelerator": 7755.275, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9474F", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB DDR5 4800MHz", - "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1793 ROUGEL: 28.8001 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 21775.5, - "metrics.result_per_accelerator": 2721.9375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 45.4712 ROUGE2: 23.1931 ROUGEL: 30.3668 TOKENS_PER_SAMPLE: 146.4 gsm8k_accuracy: 73.2 mbxp_accuracy: 60.06", + "metrics.result": 58972.6, + "metrics.result_per_accelerator": 7371.575, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "mixtral-8x7b", + "model.name": "mixtral-8x7b", + "model.number_of_parameters": 47.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9474F", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB DDR5 4800MHz", - "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 24011.7, - "metrics.result_per_accelerator": 3001.4625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5457 ROUGE2: 22.133 ROUGEL: 28.7834 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34763.2, + "metrics.result_per_accelerator": 4345.4, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9474F", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB DDR5 4800MHz", - "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1793 ROUGEL: 28.8001 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 21775.5, - "metrics.result_per_accelerator": 2721.9375, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5602 ROUGE2: 22.142 ROUGEL: 28.794 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 33023.5, + "metrics.result_per_accelerator": 4127.9375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 120, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9474F", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8580", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB DDR5 4800MHz", - "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 4096.0, + "system.memory.configuration": "32x 128GB DDR5", + "system.name": "Dell PowerEdge XE9680L (8x H200-SXM-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3545 ROUGE2: 23.2434 ROUGEL: 30.3185 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", - "metrics.result": 50970.6, - "metrics.result_per_accelerator": 6371.325, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34822.6, + "metrics.result_per_accelerator": 4352.825, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9474F", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB DDR5 4800MHz", - "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 96GB", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.351 ROUGE2: 23.2457 ROUGEL: 30.3126 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", - "metrics.result": 49864.0, - "metrics.result_per_accelerator": 6233.0, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5569 ROUGE2: 22.145 ROUGEL: 28.7929 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 33051.9, + "metrics.result_per_accelerator": 4131.4875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9474F", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "AMD", + "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 1536.0, - "system.memory.configuration": "24x 64GB DDR5 4800MHz", - "system.name": "AS-8125GS-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "16x 96GB", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", - "metrics.result": 2659.47, - "metrics.result_per_accelerator": 2659.47, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21117.0, + "metrics.result_per_accelerator": 2639.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", - "metrics.result": 2159.89, - "metrics.result_per_accelerator": 2159.89, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21536.4, + "metrics.result_per_accelerator": 2692.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", - "metrics.result": 2659.47, - "metrics.result_per_accelerator": 2659.47, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21117.0, + "metrics.result_per_accelerator": 2639.625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0794 ROUGE2: 20.1138 ROUGEL: 29.9751 GEN_LEN: 4110737", - "metrics.result": 2159.89, - "metrics.result_per_accelerator": 2159.89, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21536.4, + "metrics.result_per_accelerator": 2692.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.2", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 96.0, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 141.0, "system.accelerator.memory_config": "HBM3e", - "system.accelerator.name": "NVIDIA GH200 Grace Hopper Superchip 96GB", - "system.accelerator.total_count": 1, + "system.accelerator.name": "NVIDIA H200-SXM-141GB", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 72, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 48, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "NVIDIA Grace CPU", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "NVIDIA", - "system.interconnect.accelerator": "InfiniBand", - "system.interconnect.accelerator_host": "NVLink", - "system.memory.capacity": 480.0, - "system.memory.configuration": "15x 16DP (32GB) LPDDR5x", - "system.name": "NVIDIA GH200-GraceHopper-Superchip (1x GH200-96GB_aarch64, TensorRT)", + "system.cpu.vendor": "Intel", + "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 19808.2, - "metrics.result_per_accelerator": 2476.025, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34545.5, + "metrics.result_per_accelerator": 4318.1875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0997 ROUGE2: 20.1114 ROUGEL: 29.9739 GEN_LEN: 4114157", - "metrics.result": 19725.6, - "metrics.result_per_accelerator": 2465.7, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5036 ROUGE2: 22.0997 ROUGEL: 28.733 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 19428.4, + "metrics.result_per_accelerator": 2428.55, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 19808.2, - "metrics.result_per_accelerator": 2476.025, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34545.5, + "metrics.result_per_accelerator": 4318.1875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0997 ROUGE2: 20.1114 ROUGEL: 29.9739 GEN_LEN: 4114157", - "metrics.result": 19725.6, - "metrics.result_per_accelerator": 2465.7, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5036 ROUGE2: 22.0997 ROUGEL: 28.733 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 19428.4, + "metrics.result_per_accelerator": 2428.55, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 24180.6, - "metrics.result_per_accelerator": 3022.575, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 34822.6, + "metrics.result_per_accelerator": 4352.825, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1792 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 21888.6, - "metrics.result_per_accelerator": 2736.075, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5569 ROUGE2: 22.145 ROUGEL: 28.7929 TOKENS_PER_SAMPLE: 292.3", + "metrics.result": 33051.9, + "metrics.result_per_accelerator": 4131.4875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-SXM-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "INTEL(R) XEON(R) PLATINUM 8568Y+", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink-4", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.capacity": 1536.0, + "system.memory.configuration": "16x 96GB", + "system.name": "Dell PowerEdge XE9680 (8xH200-SXM-141GBx8, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 24180.6, - "metrics.result_per_accelerator": 3022.575, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 31149.9, + "metrics.result_per_accelerator": 3893.7375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6147 ROUGE2: 22.1792 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 21888.6, - "metrics.result_per_accelerator": 2736.075, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5583 ROUGE2: 22.1382 ROUGEL: 28.7896 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 28848.4, + "metrics.result_per_accelerator": 3606.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3632 ROUGE2: 23.2559 ROUGEL: 30.3191 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.9", - "metrics.result": 52452.0, - "metrics.result_per_accelerator": 6556.5, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 16900.1, + "metrics.result_per_accelerator": 2112.5125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3565 ROUGE2: 23.2475 ROUGEL: 30.3154 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.86", - "metrics.result": 51028.4, - "metrics.result_per_accelerator": 6378.55, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 17975.8, + "metrics.result_per_accelerator": 2246.975, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8570", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-421GE-TNHR2-LCC (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 19803.6, - "metrics.result_per_accelerator": 2475.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 16900.1, + "metrics.result_per_accelerator": 2112.5125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1015 ROUGE2: 20.1128 ROUGEL: 29.9751 GEN_LEN: 4114215", - "metrics.result": 19635.2, - "metrics.result_per_accelerator": 2454.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 17975.8, + "metrics.result_per_accelerator": 2246.975, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "gptj-99", "model.name": "gptj-99", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1017 ROUGE2: 20.1125 ROUGEL: 29.9751 GEN_LEN: 4114386", - "metrics.result": 19803.6, - "metrics.result_per_accelerator": 2475.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 31334.8, + "metrics.result_per_accelerator": 3916.85, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.1015 ROUGE2: 20.1128 ROUGEL: 29.9751 GEN_LEN: 4114215", - "metrics.result": 19635.2, - "metrics.result_per_accelerator": 2454.4, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5859 ROUGE2: 22.1562 ROUGEL: 28.8145 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 16134.4, + "metrics.result_per_accelerator": 2016.8, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 24140.1, - "metrics.result_per_accelerator": 3017.5125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 31334.8, + "metrics.result_per_accelerator": 3916.85, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6148 ROUGE2: 22.1792 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 21986.0, - "metrics.result_per_accelerator": 2748.25, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5859 ROUGE2: 22.1562 ROUGEL: 28.8145 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 16134.4, + "metrics.result_per_accelerator": 2016.8, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6153 ROUGE2: 22.1789 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 24140.1, - "metrics.result_per_accelerator": 3017.5125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.546 ROUGE2: 22.1333 ROUGEL: 28.7835 TOKENS_PER_SAMPLE: 291.5", + "metrics.result": 31149.9, + "metrics.result_per_accelerator": 3893.7375, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.6148 ROUGE2: 22.1792 ROUGEL: 28.8 TOKENS_PER_SAMPLE: 290.1", - "metrics.result": 21986.0, - "metrics.result_per_accelerator": 2748.25, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5583 ROUGE2: 22.1382 ROUGEL: 28.7896 TOKENS_PER_SAMPLE: 292.4", + "metrics.result": 28848.4, + "metrics.result_per_accelerator": 3606.05, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99.9", "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", + "system.accelerator.memory_capacity": 141.0, + "system.accelerator.memory_config": "HBM3e", + "system.accelerator.name": "NVIDIA H200-NVL-141GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 48, + "system.cpu.core_count": 192, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE7745 (8x H200-NVL-141GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3558 ROUGE2: 23.2472 ROUGEL: 30.3111 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.16 mbxp_accuracy: 59.88", - "metrics.result": 52353.8, - "metrics.result_per_accelerator": 6544.225, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5751 ROUGE2: 22.1504 ROUGEL: 28.7726 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 31216.8, + "metrics.result_per_accelerator": 3902.1, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, @@ -37895,38 +40465,41 @@ "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3545 ROUGE2: 23.2472 ROUGEL: 30.3145 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.1 mbxp_accuracy: 59.88", - "metrics.result": 50676.9, - "metrics.result_per_accelerator": 6334.6125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5724 ROUGE2: 22.1518 ROUGEL: 28.7699 TOKENS_PER_SAMPLE: 292.9", + "metrics.result": 30903.6, + "metrics.result_per_accelerator": 3862.95, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-99", + "model.name": "llama2-70b-99", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Supermicro", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, @@ -37941,406 +40514,433 @@ "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB DDR5 4800MHz", - "system.name": "SYS-821GE-TNHR (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0394 ROUGE2: 20.0943 ROUGEL: 29.9329 GEN_LEN: 4100159", - "metrics.result": 19711.1, - "metrics.result_per_accelerator": 2463.8875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20663.7, + "metrics.result_per_accelerator": 2582.9625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": null, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "InfiniBand", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.039 ROUGE2: 20.0945 ROUGEL: 29.933 GEN_LEN: 4100192", - "metrics.result": 19233.8, - "metrics.result_per_accelerator": 2404.225, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21134.4, + "metrics.result_per_accelerator": 2641.8, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99", - "model.name": "gptj-99", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": null, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "InfiniBand", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5768 ROUGE2: 22.1293 ROUGEL: 28.7645 TOKENS_PER_SAMPLE: 292.1", - "metrics.result": 24459.6, - "metrics.result_per_accelerator": 3057.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 20663.7, + "metrics.result_per_accelerator": 2582.9625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": null, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "InfiniBand", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5777 ROUGE2: 22.1298 ROUGEL: 28.7648 TOKENS_PER_SAMPLE: 292.1", - "metrics.result": 21327.0, - "metrics.result_per_accelerator": 2665.875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.1118 ROUGE2: 20.1071 ROUGEL: 29.9676 GEN_LEN: 4153194", + "metrics.result": 21134.4, + "metrics.result_per_accelerator": 2641.8, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": null, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "InfiniBand", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5768 ROUGE2: 22.1293 ROUGEL: 28.7645 TOKENS_PER_SAMPLE: 292.1", - "metrics.result": 24459.6, - "metrics.result_per_accelerator": 3057.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5767 ROUGE2: 22.1512 ROUGEL: 28.773 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 30989.5, + "metrics.result_per_accelerator": 3873.6875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": null, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "InfiniBand", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5777 ROUGE2: 22.1298 ROUGEL: 28.7648 TOKENS_PER_SAMPLE: 292.1", - "metrics.result": 21327.0, - "metrics.result_per_accelerator": 2665.875, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5968 ROUGE2: 22.1606 ROUGEL: 28.7638 TOKENS_PER_SAMPLE: 293.1", + "metrics.result": 12979.4, + "metrics.result_per_accelerator": 1622.425, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", + "model.mlperf_name": "llama2-70b-interactive-99", + "model.name": "llama2-70b-interactive-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": null, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "InfiniBand", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3664 ROUGE2: 23.2584 ROUGEL: 30.3279 TOKENS_PER_SAMPLE: 145.9 gsm8k_accuracy: 73.08 mbxp_accuracy: 59.86", - "metrics.result": 52817.7, - "metrics.result_per_accelerator": 6602.2125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.643467134956637 exact_match: 90.07990403706154 TOKENS_PER_SAMPLE: 645.6", + "metrics.result": 397.31, + "metrics.result_per_accelerator": 49.66375, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": null, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "InfiniBand", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 45.3528 ROUGE2: 23.2465 ROUGEL: 30.3132 TOKENS_PER_SAMPLE: 146.0 gsm8k_accuracy: 73.08 mbxp_accuracy: 59.9", - "metrics.result": 50797.2, - "metrics.result_per_accelerator": 6349.65, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGEL: 21.578037465211967 exact_match: 90.02991065519522 TOKENS_PER_SAMPLE: 646.3", + "metrics.result": 277.125, + "metrics.result_per_accelerator": 34.640625, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama3.1-405b", + "model.name": "llama3_1-405b", + "model.number_of_parameters": 405.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", "submission.division": "closed", - "submission.organization": "Sustainable_Metal_Cloud", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, - "system.accelerator.memory_capacity": null, + "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 32, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", "system.interconnect.accelerator": "NVLink", - "system.interconnect.accelerator_host": "InfiniBand", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, "system.memory.configuration": "32x 64GB DDR5", - "system.name": "SMC H100 (8x H100-SXM-80GB, TensorRT)", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0343 ROUGE2: 20.0912 ROUGEL: 29.9254 GEN_LEN: 4102192", - "metrics.result": 19889.7, - "metrics.result_per_accelerator": 2486.2125, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5767 ROUGE2: 22.1512 ROUGEL: 28.773 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 30989.5, + "metrics.result_per_accelerator": 3873.6875, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "HPE", + "submission.division": "closed", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, @@ -38348,45 +40948,48 @@ "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, + "system.cpu.frequency": null, "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.0346 ROUGE2: 20.0913 ROUGEL: 29.9259 GEN_LEN: 4102422", - "metrics.result": 19627.6, - "metrics.result_per_accelerator": 2453.45, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5968 ROUGE2: 22.1606 ROUGEL: 28.7638 TOKENS_PER_SAMPLE: 293.1", + "metrics.result": 12979.4, + "metrics.result_per_accelerator": 1622.425, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "gptj-99.9", - "model.name": "gptj-99.9", - "model.number_of_parameters": 6.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", + "model.mlperf_name": "llama2-70b-interactive-99.9", + "model.name": "llama2-70b-interactive-99.9", + "model.number_of_parameters": 70.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "22.04.4 LTS", + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "HPE", + "submission.division": "closed", + "submission.organization": "Dell", "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, @@ -38394,413 +40997,342 @@ "system.accelerator.name": "NVIDIA H100-SXM-80GB", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 4.5 MiB (96 instances), L1i: 3 MiB (96 instances), L2: 192 MiB (96 instances), L3: 210 MiB (2 instances)", + "system.cpu.caches": null, "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 3.8, + "system.cpu.frequency": null, "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "HPE Cray XD670 (8x H100-SXM-80GB, TensorRT)", - "system.number_of_nodes": 1, - "system.type": "datacenter" - }, - { - "metrics.accuracy": "ROUGE1: 45.4026 ROUGE2: 23.228 ROUGEL: 30.3021 TOKENS_PER_SAMPLE: 145.3 gsm8k_accuracy: 73.64 mbxp_accuracy: 59.98", - "metrics.result": 18839.0, - "metrics.result_per_accelerator": 4709.75, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", - "software.framework.ai": null, - "software.framework.cuda": "12.4", - "software.framework.cudnn": null, - "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", - "submission.availability": "available", - "submission.division": "open", - "submission.organization": "HPE", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB, L1i: 2 MiB, L2: 128 MiB (64 instances), L3: 320 MiB (2 instances)", - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB", - "system.name": "HPE ProLiant DL380a Gen11 (4x H100-NVL-94GB, TensorRT)", - "system.number_of_nodes": 1, - "system.type": "datacenter" - }, - { - "metrics.accuracy": "ROUGE1: 45.4098 ROUGE2: 23.2338 ROUGEL: 30.3109 TOKENS_PER_SAMPLE: 145.3 gsm8k_accuracy: 73.66 mbxp_accuracy: 60.0", - "metrics.result": 17453.4, - "metrics.result_per_accelerator": 4363.35, - "metrics.units": "Tokens/s", - "model.architecture": "LLM", - "model.mlperf_name": "mixtral-8x7b", - "model.name": "mixtral-8x7b", - "model.number_of_parameters": 47.0, - "model.weight_data_types": "fp16", - "software.framework": "TensorRT 10.2.0, CUDA 12.4", - "software.framework.ai": null, - "software.framework.cuda": "12.4", - "software.framework.cudnn": null, - "software.framework.jetpack": null, - "software.framework.pytorch": null, - "software.framework.rocm": null, - "software.framework.sima": null, - "software.framework.tensorrt": "10.2.0", - "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4 LTS", - "submission.availability": "available", - "submission.division": "open", - "submission.organization": "HPE", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 94.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-NVL-94GB", - "system.accelerator.total_count": 4, - "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d: 3 MiB, L1i: 2 MiB, L2: 128 MiB (64 instances), L3: 320 MiB (2 instances)", - "system.cpu.core_count": 32, - "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4.0, - "system.cpu.model": "INTEL(R) XEON(R) GOLD 6530", - "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 1024.0, - "system.memory.configuration": "16x 64GB", - "system.name": "HPE ProLiant DL380a Gen11 (4x H100-NVL-94GB, TensorRT)", + "system.interconnect.accelerator": "NVLink", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.5491 ROUGE2: 22.1761 ROUGEL: 28.7877 TOKENS_PER_SAMPLE: 293.9", - "metrics.result": 1468.24, - "metrics.result_per_accelerator": 367.06, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5751 ROUGE2: 22.1504 ROUGEL: 28.7726 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 31216.8, + "metrics.result_per_accelerator": 3902.1, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, "model.weight_data_types": "fp8", - "software.framework": "vLLM 0.5.2", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.5.2", - "software.operating_system": "Ubuntu 22.04 (linux-6.5.0-35-generic-glibc2.35)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "NeuralMagic", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, + "submission.division": "closed", + "submission.organization": "Dell", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d cache: 3 MiB (64 instances), L1i cache: 2 MiB (64 instances), L2 cache: 128 MiB (64 instances), L3 cache: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4100.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": null, - "system.memory.configuration": "undefined", - "system.name": "SYS-821GE-TNHR H100 'beaker' (4x H100-SXM-80GB, vLLM, FP8)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.2003 ROUGE2: 21.8589 ROUGEL: 28.4858 TOKENS_PER_SAMPLE: 586.9", - "metrics.result": 1577.11, - "metrics.result_per_accelerator": 394.2775, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5724 ROUGE2: 22.1518 ROUGEL: 28.7699 TOKENS_PER_SAMPLE: 292.9", + "metrics.result": 30903.6, + "metrics.result_per_accelerator": 3862.95, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "int4", - "software.framework": "vLLM 0.5.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.5.2", - "software.operating_system": "Ubuntu 22.04 (linux-6.5.0-35-generic-glibc2.35)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 22.04.4", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "NeuralMagic", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 4, + "submission.division": "closed", + "submission.organization": "Dell", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 80.0, "system.accelerator.memory_config": "HBM3", "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d cache: 3 MiB (64 instances), L1i cache: 2 MiB (64 instances), L2 cache: 128 MiB (64 instances), L3 cache: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 48, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4100.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.frequency": null, + "system.cpu.model": "Intel(R) Xeon(R) Platinum 8468", "system.cpu.vcpu_count": null, "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.interconnect.accelerator": "NVLink", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": null, - "system.memory.configuration": "undefined", - "system.name": "SYS-821GE-TNHR H100 'beaker' (4x H100-SXM-80GB, vLLM, GPTQ)", + "system.memory.capacity": 2048.0, + "system.memory.configuration": "32x 64GB DDR5", + "system.name": "Dell PowerEdge XE9680 (8x H100-SXM-80GB, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.1657 ROUGE2: 21.8046 ROUGEL: 28.4386 TOKENS_PER_SAMPLE: 291.2", - "metrics.result": 1164.04, - "metrics.result_per_accelerator": 291.01, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5645 ROUGE2: 22.1587 ROUGEL: 28.7857 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 3481.53, + "metrics.result_per_accelerator": 435.19125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "int4", - "software.framework": "vLLM 0.5.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.5.2", - "software.operating_system": "Ubuntu 22.04 (linux-6.5.0-35-generic-glibc2.35)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "NeuralMagic", - "submission.scenario": "Server", - "system.accelerator.count_per_node": 4, - "system.accelerator.memory_capacity": 80.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H100-SXM-80GB", - "system.accelerator.total_count": 4, + "submission.division": "closed", + "submission.organization": "Dell", + "submission.scenario": "Offline", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d cache: 3 MiB (64 instances), L1i cache: 2 MiB (64 instances), L2 cache: 128 MiB (64 instances), L3 cache: 120 MiB (2 instances)", - "system.cpu.core_count": 32, + "system.cpu.caches": null, + "system.cpu.core_count": 384, "system.cpu.count_per_node": 2, - "system.cpu.frequency": 4100.0, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8462Y+", + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": null, - "system.memory.configuration": "undefined", - "system.name": "SYS-821GE-TNHR H100 'beaker' (4x H100-SXM-80GB, vLLM, GPTQ)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB", + "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.0047 ROUGE2: 21.7276 ROUGEL: 28.4008 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 424.895, - "metrics.result_per_accelerator": 212.4475, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5496 ROUGE2: 22.1516 ROUGEL: 28.7732 TOKENS_PER_SAMPLE: 294.1", + "metrics.result": 3201.13, + "metrics.result_per_accelerator": 400.14125, "metrics.units": "Tokens/s", "model.architecture": "LLM", "model.mlperf_name": "llama2-70b-99", "model.name": "llama2-70b-99", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp32", - "software.framework": "vLLM 0.5.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.5.2", - "software.operating_system": "Ubuntu 23.04 (linux-6.2.0-39-generic-glibc2.37)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "NeuralMagic", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 23.64703369140625, - "system.accelerator.memory_config": null, - "system.accelerator.name": "NVIDIA GeForce RTX 4090", - "system.accelerator.total_count": 2, + "submission.division": "closed", + "submission.organization": "Dell", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d cache: 1.1 MiB (24 instances), L1i cache: 768 KiB (24 instances), L2 cache: 48 MiB (24 instances), L3 cache: 45 MiB (1 instance)", - "system.cpu.core_count": 24, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": 4800.0, - "system.cpu.model": "Intel(R) Xeon(R) w7-2495X", + "system.cpu.caches": null, + "system.cpu.core_count": 384, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": null, - "system.memory.capacity": null, - "system.memory.configuration": "undefined", - "system.name": "GATE Overflow Intel Sapphire Rapids RTX 4090 (2x RTX 4090, vLLM, FP8)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB", + "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.0047 ROUGE2: 21.7276 ROUGEL: 28.4008 TOKENS_PER_SAMPLE: 291.9", - "metrics.result": 424.895, - "metrics.result_per_accelerator": 212.4475, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0584 ROUGE2: 20.1404 ROUGEL: 29.9742 GEN_LEN: 4125446", + "metrics.result": 6946.68, + "metrics.result_per_accelerator": 868.335, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp32", - "software.framework": "vLLM 0.5.2", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.5.2", - "software.operating_system": "Ubuntu 23.04 (linux-6.2.0-39-generic-glibc2.37)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "NeuralMagic", + "submission.division": "closed", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 2, - "system.accelerator.memory_capacity": 23.64703369140625, - "system.accelerator.memory_config": null, - "system.accelerator.name": "NVIDIA GeForce RTX 4090", - "system.accelerator.total_count": 2, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d cache: 1.1 MiB (24 instances), L1i cache: 768 KiB (24 instances), L2 cache: 48 MiB (24 instances), L3 cache: 45 MiB (1 instance)", - "system.cpu.core_count": 24, - "system.cpu.count_per_node": 1, - "system.cpu.frequency": 4800.0, - "system.cpu.model": "Intel(R) Xeon(R) w7-2495X", + "system.cpu.caches": null, + "system.cpu.core_count": 384, + "system.cpu.count_per_node": 2, + "system.cpu.frequency": null, + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": null, - "system.memory.capacity": null, - "system.memory.configuration": "undefined", - "system.name": "GATE Overflow Intel Sapphire Rapids RTX 4090 (2x RTX 4090, vLLM, FP8)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB", + "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.1411 ROUGE2: 21.8139 ROUGEL: 28.4395 TOKENS_PER_SAMPLE: 292.9", - "metrics.result": 923.333, - "metrics.result_per_accelerator": 115.416625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.055 ROUGE2: 20.1229 ROUGEL: 29.9616 GEN_LEN: 4127336", + "metrics.result": 6213.34, + "metrics.result_per_accelerator": 776.6675, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp32", - "software.framework": "vLLM 0.5.2", + "model.mlperf_name": "gptj-99.9", + "model.name": "gptj-99.9", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.5.2", - "software.operating_system": "Ubuntu 22.04 (linux-5.15.0-94-generic-glibc2.35)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "NeuralMagic", - "submission.scenario": "Offline", + "submission.division": "closed", + "submission.organization": "Dell", + "submission.scenario": "Server", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 48.0, "system.accelerator.memory_config": "GDDR6", "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d cache: 1.3 MiB (40 instances), L1i cache: 1.3 MiB (40 instances), L2 cache: 40 MiB (40 instances), L3 cache: 160 MiB (5 instances)", - "system.cpu.core_count": 4, + "system.cpu.caches": null, + "system.cpu.core_count": 384, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9254 24-Core Processor", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": null, - "system.memory.configuration": "undefined", - "system.name": "Crusoe Cloud L40S (8x L40S PCIe, vLLM)", + "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB", + "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.1411 ROUGE2: 21.8139 ROUGEL: 28.4395 TOKENS_PER_SAMPLE: 292.9", - "metrics.result": 923.333, - "metrics.result_per_accelerator": 115.416625, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.0584 ROUGE2: 20.1404 ROUGEL: 29.9742 GEN_LEN: 4125446", + "metrics.result": 6946.68, + "metrics.result_per_accelerator": 868.335, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99.9", - "model.name": "llama2-70b-99.9", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp32", - "software.framework": "vLLM 0.5.2", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": null, + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, - "software.framework.vllm": "0.5.2", - "software.operating_system": "Ubuntu 22.04 (linux-5.15.0-94-generic-glibc2.35)", + "software.framework.tensorrt": "10.8", + "software.framework.vllm": null, + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "NeuralMagic", + "submission.division": "closed", + "submission.organization": "Dell", "submission.scenario": "Offline", "system.accelerator.count_per_node": 8, "system.accelerator.memory_capacity": 48.0, @@ -38808,157 +41340,166 @@ "system.accelerator.name": "NVIDIA L40S", "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", - "system.cpu.caches": "L1d cache: 1.3 MiB (40 instances), L1i cache: 1.3 MiB (40 instances), L2 cache: 40 MiB (40 instances), L3 cache: 160 MiB (5 instances)", - "system.cpu.core_count": 4, + "system.cpu.caches": null, + "system.cpu.core_count": 384, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "AMD EPYC 9254 24-Core Processor", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, "system.cpu.vendor": "AMD", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": "PCIe", - "system.memory.capacity": null, - "system.memory.configuration": "undefined", - "system.name": "Crusoe Cloud L40S (8x L40S PCIe, vLLM)", + "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB", + "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 42.9783 ROUGE2: 24.3368 ROUGEL: 31.4041 TOKENS_PER_SAMPLE: 164.6", - "metrics.result": 11189.1, - "metrics.result_per_accelerator": 11189.1, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 43.055 ROUGE2: 20.1229 ROUGEL: 29.9616 GEN_LEN: 4127336", + "metrics.result": 6213.34, + "metrics.result_per_accelerator": 776.6675, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", - "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "CUDA 12.4", + "model.mlperf_name": "gptj-99", + "model.name": "gptj-99", + "model.number_of_parameters": 6.0, + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "NVIDIA", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 1, + "submission.division": "closed", + "submission.organization": "Dell", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 384, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (1x H200-SXM-141GB)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB", + "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 44.46 ROUGE2: 22.4526 ROUGEL: 29.9493 TOKENS_PER_SAMPLE: 234.4", - "metrics.result": 4575.06, - "metrics.result_per_accelerator": 4575.06, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5645 ROUGE2: 22.1587 ROUGEL: 28.7857 TOKENS_PER_SAMPLE: 291.9", + "metrics.result": 3481.53, + "metrics.result_per_accelerator": 435.19125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "CUDA 12.4", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.4", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": null, + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Ubuntu 22.04.4", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "NVIDIA", + "submission.division": "closed", + "submission.organization": "Dell", "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": 141.0, - "system.accelerator.memory_config": "HBM3", - "system.accelerator.name": "NVIDIA H200-SXM-141GB", - "system.accelerator.total_count": 1, + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 56, + "system.cpu.core_count": 384, "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "Intel(R) Xeon(R) Platinum 8480C", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "Intel", - "system.interconnect.accelerator": "NVLink-4", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "PCIe-5", "system.interconnect.accelerator_host": "PCIe-5", - "system.memory.capacity": 2048.0, - "system.memory.configuration": "32x 64GB MTC40F2046S1RC48BA1", - "system.name": "NVIDIA H200 (1x H200-SXM-141GB)", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB", + "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" }, { - "metrics.accuracy": "ROUGE1: 43.2263 ROUGE2: 24.3279 ROUGEL: 31.3816 TOKENS_PER_SAMPLE: 166.3", - "metrics.result": 184.893, - "metrics.result_per_accelerator": 184.893, + "benchmark.name": "mlperf-inference", + "benchmark.version": "5.0", + "datetime": "2025-04-03T22:56:53", + "metrics.accuracy": "ROUGE1: 44.5496 ROUGE2: 22.1516 ROUGEL: 28.7732 TOKENS_PER_SAMPLE: 294.1", + "metrics.result": 3201.13, + "metrics.result_per_accelerator": 400.14125, "metrics.units": "Tokens/s", "model.architecture": "LLM", - "model.mlperf_name": "llama2-70b-99", - "model.name": "llama2-70b-99", + "model.mlperf_name": "llama2-70b-99.9", + "model.name": "llama2-70b-99.9", "model.number_of_parameters": 70.0, - "model.weight_data_types": "fp16", - "software.framework": "Jetpack 6.0, TensorRT 10.1, CUDA 12.2", + "model.weight_data_types": "fp8", + "software.framework": "TensorRT 10.8, CUDA 12.8", "software.framework.ai": null, - "software.framework.cuda": "12.2", + "software.framework.cuda": "12.8", "software.framework.cudnn": null, - "software.framework.jetpack": "6.0", + "software.framework.jetpack": null, "software.framework.pytorch": null, "software.framework.rocm": null, "software.framework.sima": null, - "software.framework.tensorrt": "10.1", + "software.framework.tensorrt": "10.8", "software.framework.vllm": null, - "software.operating_system": "Jetson r36.3.1 L4T", + "software.operating_system": "Ubuntu 24.04", "submission.availability": "available", - "submission.division": "open", - "submission.organization": "NVIDIA", - "submission.scenario": "Offline", - "system.accelerator.count_per_node": 1, - "system.accelerator.memory_capacity": null, - "system.accelerator.memory_config": "LPDDR5", - "system.accelerator.name": "NVIDIA Jetson AGX Orin 64G", - "system.accelerator.total_count": 1, + "submission.division": "closed", + "submission.organization": "Dell", + "submission.scenario": "Server", + "system.accelerator.count_per_node": 8, + "system.accelerator.memory_capacity": 48.0, + "system.accelerator.memory_config": "GDDR6", + "system.accelerator.name": "NVIDIA L40S", + "system.accelerator.total_count": 8, "system.accelerator.vendor": "NVIDIA", "system.cpu.caches": null, - "system.cpu.core_count": 12, - "system.cpu.count_per_node": 1, + "system.cpu.core_count": 384, + "system.cpu.count_per_node": 2, "system.cpu.frequency": null, - "system.cpu.model": "12-core ARM Cortex-A78AE CPU", + "system.cpu.model": "AMD EPYC 9965 192-Core Processor", "system.cpu.vcpu_count": null, - "system.cpu.vendor": "ARM", - "system.interconnect.accelerator": null, - "system.interconnect.accelerator_host": null, - "system.memory.capacity": 64.0, - "system.memory.configuration": "64GB 256-bit LPDDR5", - "system.name": "NVIDIA Jetson AGX Orin Developer Kit 64G (TensorRT)", + "system.cpu.vendor": "AMD", + "system.interconnect.accelerator": "PCIe-5", + "system.interconnect.accelerator_host": "PCIe-5", + "system.memory.capacity": 3072.0, + "system.memory.configuration": "24x 128GB", + "system.name": "Dell PowerEdge XE7745 (8x L40S, TensorRT)", "system.number_of_nodes": 1, "system.type": "datacenter" } -] \ No newline at end of file +]