{ "query": { "task": "LLM serving (MLPerf datacenter)" }, "task": "LLM serving (MLPerf datacenter)", "size_class": "LLM >40B", "batch": "unknown", "mixed": false, "mixed_warning": null, "groups": [ { "size": "LLM >40B", "batch": "unknown", "platforms": 93, "rows": 372 }, { "size": "LLM 5–14B", "batch": "unknown", "platforms": 48, "rows": 136 }, { "size": "unknown", "batch": "unknown", "platforms": 32, "rows": 97 } ], "notes": [], "count": 93, "results": [ { "platform": "Crusoe Cloud 64x8 MI355X (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/crusoe-cloud-64x8-mi355x-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 5749440, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "mxfp4", "stack": "vLLM 0.22.1 (ROCm 7.2.2), PyTorch 2.10", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Crusoe", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system 512xMI355X_Crusoe_GPT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Crusoe/results/512xMI355X_Crusoe_GPT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 1, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "CoreWeave GB300 NVL72 (4x NVIDIA GB300)", "platform_url": "/platform/coreweave-gb300-nvl72-4x-nvidia-gb300.html", "perf": 1197710, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentML:mlperf-inf-mm-q3vl-v6.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "CoreWeave", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system GB300-NVL72_GB300-288GB_aarch64x72_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/CoreWeave/results/GB300-NVL72_GB300-288GB_aarch64x72_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.2083, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Nebius GB300 NVL72 (4x NVIDIA GB300)", "platform_url": "/platform/nebius-gb300-nvl72-4x-nvidia-gb300.html", "perf": 1186750, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Nebius", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system GB300-NVL72_GB300-288GB_aarch64x72_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Nebius/results/GB300-NVL72_GB300-288GB_aarch64x72_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.2064, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "NVIDIA GB300 NVL72 (4x NVIDIA GB300)", "platform_url": "/platform/nvidia-gb300-nvl72-4x-nvidia-gb300.html", "perf": 1126850, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "NVIDIA", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0078", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/NVIDIA/results/GB300-NVL72_GB300-288GB_aarch64x72_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.196, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "AMD MI355X Custom Cluster (8x AMD Instinct MI355X 288GB HBM3e (x87))", "platform_url": "/platform/amd-mi355x-custom-cluster-8x-amd-instinct-mi355x-288gb-hbm3e-x87.html", "perf": 1042110, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "PyTorch 2.9.1+git8907517, ROCm 7.0.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "AMD", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0001", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/AMD/results/87xMI355X_22xEPYC_9575F", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.1813, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "9x8xMI355X_2xEPYC_9575F (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/9x8xmi355x-2xepyc-9575f-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 1039900, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "mxfp4", "stack": "vLLM 0.22.1.dev0, PyTorch 2.10.0, ROCm 7.2.2", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "AMD", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system 9x8xMI355X_2xEPYC_9575F", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/AMD/results/9x8xMI355X_2xEPYC_9575F", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.1809, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "AMD MI355X Custom Cluster (8x AMD Instinct MI355X 288GB HBM3e (x94))", "platform_url": "/platform/amd-mi355x-custom-cluster-8x-amd-instinct-mi355x-288gb-hbm3e-x94.html", "perf": 1031070, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "PyTorch 2.9.1+git8907517, ROCm 7.0.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "AMD", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0003", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/AMD/results/94xMI355X_24xEPYC_9575F", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.1793, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "CoreWeave GB200 NVL72 (4x NVIDIA GB200)", "platform_url": "/platform/coreweave-gb200-nvl72-4x-nvidia-gb200.html", "perf": 911566, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "CoreWeave", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system GB200-NVL72_GB200-186GB_aarch64x72_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/CoreWeave/results/GB200-NVL72_GB200-186GB_aarch64x72_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.1585, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "NVIDIA GB200 NVL72 (4x NVIDIA GB200)", "platform_url": "/platform/nvidia-gb200-nvl72-4x-nvidia-gb200.html", "perf": 899218, "perf_unit": "tok/s", "metric": "throughput_server", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "NVIDIA", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0075", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/NVIDIA/results/GB200-NVL72_GB200-186GB_aarch64x72_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.1564, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Heterogeneous Four-Cluster (8x AMD Instinct MI300X 192GB HBM3 (x2x8) and AMD Instinct MI355X 288GB HBM3e (x2x8))", "platform_url": "/platform/heterogeneous-four-cluster-8x-amd-instinct-mi300x-192gb-hbm3-x2x8-and-amd-instin.html", "perf": 285454, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp8", "stack": "ROCm", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell_MangoBoost", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system 2x8xMI355X_2x8xMI300X", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Dell_MangoBoost/results/2x8xMI355X_2x8xMI300X", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0496, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "MangoBoost 2x8 MI355X (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/mangoboost-2x8-mi355x-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 228488, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp8", "stack": "ROCm 7.2.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "MangoBoost", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system 16xMI355X_2xEPYC_9575F", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/MangoBoost/results/16xMI355X_2xEPYC_9575F", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0397, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Cisco_UCS_B300x16_G200_4x2_RO (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/cisco-ucs-b300x16-g200-4x2-ro-8x-nvidia-b300-sxm-270gb.html", "perf": 227570, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14.1.48; TensorRT-LLM 1.3.0rc0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Cisco", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system Cisco_UCS_B300x16_G200_4x2_RO", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Cisco/results/Cisco_UCS_B300x16_G200_4x2_RO", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0396, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Cisco UCS 16xMI350X (8x AMD Instinct MI350X 288GB HBM3e)", "platform_url": "/platform/cisco-ucs-16xmi350x-8x-amd-instinct-mi350x-288gb-hbm3e.html", "perf": 159932, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "vLLM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Cisco", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0011", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Cisco/results/Cisco_UCS_16xMI350x", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0278, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Dell_MangoBoost Heterogeneous Cluster (8x AMD Instinct MI300X 192GB HBM3 (x8) and AMD Instinct MI325X 256GB HBM3e (x8) and AMD Instinct MI355X 288GB HBM3e (x8))", "platform_url": "/platform/dell-mangoboost-heterogeneous-cluster-8x-amd-instinct-mi300x-192gb-hbm3-x8-and-a.html", "perf": 151843, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp8", "stack": "ROCm", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell_MangoBoost", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0031", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Dell_MangoBoost/results/8xMI300X_2xEPYC_9534_8xMI325X_2xEPYC_9655_8xMI355X_2xEPYC_9965", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0264, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Orrick Optimized 6x NVIDIA B200 L2 (6x NVIDIA B200-SXM-180GB)", "platform_url": "/platform/orrick-optimized-6x-nvidia-b200-l2-6x-nvidia-b200-sxm-180gb.html", "perf": 130928, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "nvfp4", "stack": "TensorRT-LLM 1.3.0rc20", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "B", "confidence_label": "B – documented third-party method", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Orrick Industries LLC", "source": { "name": "MLPerf Inference v6.1 (open, datacenter) system b200_l2_6x", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/open/Orrick Industries LLC/results/b200_l2_6x", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0228, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "smci355-ccs-aus-m13-13 (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/smci355-ccs-aus-m13-13-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 121818, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "PyTorch 2.10.0+git8514f05, ROCm 7.2.2", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "AMD", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system 8xMI355X_2xEPYC_9575F", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/AMD/results/8xMI355X_2xEPYC_9575F", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0212, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Dell XE9785L_8xMI355X (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/dell-xe9785l-8xmi355x-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 117804, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "vLLM 0.22.0, PyTorch 2.10.0, ROCm 7.2.2", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system XE9785L_8xMI355X", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Dell/results/XE9785L_8xMI355X", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0205, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "G4826 (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/g4826-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 118042, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "vLLM 0.22.0, PyTorch 2.10.0+git8514f05, ROCm 7.2.53211", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "MiTAC", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system 8xMI355X_2xEPYC-9755", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/MiTAC/results/8xMI355X_2xEPYC-9755", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0205, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "HPE ProLiant Compute XD685 (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/hpe-proliant-compute-xd685-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 117355, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "vLLM 0.22.0, PyTorch 2.10.0, ROCm 7.2.2", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "HPE", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system HPE_XD685_MI355X_288GBx8_vLLM", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/HPE/results/HPE_XD685_MI355X_288GBx8_vLLM", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0204, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "AS-4126GS-NMR-LCC (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/as-4126gs-nmr-lcc-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 117005, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "vLLM 0.22.0, PyTorch 2.10.0, ROCm 7.2.4", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Supermicro", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system 8xMI355X_2xEPYC-9575F", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Supermicro/results/8xMI355X_2xEPYC-9575F", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0204, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "PowerEdge XE9785L (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/poweredge-xe9785l-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 116953, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp8", "stack": "ROCm", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell_MangoBoost", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system XE9785L", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Dell_MangoBoost/results/XE9785L", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0203, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "CoreWeave B300 SXM (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/coreweave-b300-sxm-8x-nvidia-b300-sxm-270gb.html", "perf": 115530, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentML:mlperf-inf-mm-q3vl-v6.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "CoreWeave", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system B300-SXM-270GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/CoreWeave/results/B300-SXM-270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0201, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "PowerEdge XE9785 (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/poweredge-xe9785-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 115294, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp8", "stack": "ROCm", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell_MangoBoost", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system XE9785", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Dell_MangoBoost/results/XE9785", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0201, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "8xMI355X_2xEPYC_9575F (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/8xmi355x-2xepyc-9575f-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 115567, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "vLLM 0.22.0, PyTorch 2.10.0, ROCm 7.2.2", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Oracle", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system 8xMI355X_2xEPYC-9575F", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Oracle/results/8xMI355X_2xEPYC-9575F", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0201, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Cisco UCS C880A M8 (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/cisco-ucs-c880a-m8-8x-nvidia-b300-sxm-270gb.html", "perf": 115730, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.13, CUDA 13.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Cisco", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0010", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Cisco/results/C880A_M8_B300_288GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0201, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "G894-AD3-AAX7 (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/g894-ad3-aax7-8x-nvidia-b300-sxm-270gb.html", "perf": 115034, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "GigaComputing", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system G894-AD3-AAX7_B300-SXM-270GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/GigaComputing/results/G894-AD3-AAX7_B300-SXM-270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.02, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "G894-ZD3-AAX7 (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/g894-zd3-aax7-8x-nvidia-b300-sxm-270gb.html", "perf": 114826, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "GigaComputing", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system G894-ZD3-AAX7_B300-SXM-270GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/GigaComputing/results/G894-ZD3-AAX7_B300-SXM-270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.02, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "HPE Compute XD690 (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/hpe-compute-xd690-8x-nvidia-b300-sxm-270gb.html", "perf": 114569, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.2", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "HPE", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system HPE_XD690_B300_SXM_270GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/HPE/results/HPE_XD690_B300_SXM_270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0199, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "QuantaGrid D75H-10U (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/quantagrid-d75h-10u-8x-nvidia-b300-sxm-270gb.html", "perf": 113985, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Quanta_Cloud_Technology", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system D75H-10U_B300-270GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Quanta_Cloud_Technology/results/D75H-10U_B300-270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0198, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "XA NB3I-E12 (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/xa-nb3i-e12-8x-nvidia-b300-sxm-270gb.html", "perf": 113455, "perf_unit": "tok/s", "metric": "throughput_server", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "ASUSTeK", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system XA_NB3I-E12_B300x8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/ASUSTeK/results/XA_NB3I-E12_B300x8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0197, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Dell XE9785L (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/dell-xe9785l-8x-nvidia-b300-sxm-270gb.html", "perf": 113101, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system XE9785L_x86_B300-SXM-270GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Dell/results/XE9785L_x86_B300-SXM-270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0197, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "NVIDIA DGX B300 (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/nvidia-dgx-b300-8x-nvidia-b300-sxm-270gb.html", "perf": 112954, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "NVIDIA", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0073", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/NVIDIA/results/B300-SXM-270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0196, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "AS-8126GS-NB3RT (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/as-8126gs-nb3rt-8x-nvidia-b300-sxm-270gb.html", "perf": 110802, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Supermicro", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system AS-8126GS-NB3RT_B300-SXM-288GB_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Supermicro/results/AS-8126GS-NB3RT_B300-SXM-288GB_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0193, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "SYS-822GS-NB3RT (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/sys-822gs-nb3rt-8x-nvidia-b300-sxm-270gb.html", "perf": 111247, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Supermicro", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system SYS-822GS-NB3RT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Supermicro/results/SYS-822GS-NB3RT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0193, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Inventec P5800G7 (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/inventec-p5800g7-8x-nvidia-b300-sxm-270gb.html", "perf": 109933, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Inventec", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system P5800G7-B300-SXM-270GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Inventec/results/P5800G7-B300-SXM-270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0191, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Inventec P9000G7 (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/inventec-p9000g7-8x-nvidia-b300-sxm-270gb.html", "perf": 109755, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Inventec", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system P9000G7-B300-SXM-270GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Inventec/results/P9000G7-B300-SXM-270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0191, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Dell XE9780 (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/dell-xe9780-8x-nvidia-b300-sxm-270gb.html", "perf": 109257, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, Dell Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentML:", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system XE9780-B300-SXM-270GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Dell/results/XE9780-B300-SXM-270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.019, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "G894-SD3-AAX7 (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/g894-sd3-aax7-8x-nvidia-b300-sxm-270gb.html", "perf": 109343, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "GigaComputing", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0039", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/GigaComputing/results/G894-SD3-AAX7_B300-SXM-270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.019, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Lenovo ThinkSystem SR680a V4 (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/lenovo-thinksystem-sr680a-v4-8x-nvidia-b300-sxm-270gb.html", "perf": 109141, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Lenovo", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0066", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Lenovo/results/SR680aV4", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.019, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Nebius B300 n1 (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/nebius-b300-n1-8x-nvidia-b300-sxm-270gb.html", "perf": 108529, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Nebius", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system B300-SXM-270GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Nebius/results/B300-SXM-270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0189, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Dell PowerEdge XE9780L (8x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/dell-poweredge-xe9780l-8x-nvidia-b300-sxm-270gb.html", "perf": 108669, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0024", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Dell/results/XE9780L-B300-SXM-270GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0189, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "HPE ProLiant Compute XD685 (8x NVIDIA B200-SXM-180GB)", "platform_url": "/platform/hpe-proliant-compute-xd685-8x-nvidia-b200-sxm-180gb.html", "perf": 103954, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.11, CUDA 13.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "HPE", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0048", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/HPE/results/HPE_ProLiant_XD685_B200_SXM_192GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0181, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "smci355-ccs-aus-m09-17 (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/smci355-ccs-aus-m09-17-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 103480, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "PyTorch 2.9.1+git8907517, ROCm 7.0.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "AMD", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0002", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/AMD/results/8xMI355X_2xEPYC_9575F", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.018, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "CoreWeave B200 SXM (8x NVIDIA B200-SXM-180GB)", "platform_url": "/platform/coreweave-b200-sxm-8x-nvidia-b200-sxm-180gb.html", "perf": 102703, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "CoreWeave", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system B200-SXM-180GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/CoreWeave/results/B200-SXM-180GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0179, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "SYS-822GS-NBRT (8x NVIDIA B200-SXM-180GB)", "platform_url": "/platform/sys-822gs-nbrt-8x-nvidia-b200-sxm-180gb.html", "perf": 101595, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Supermicro", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0099", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Supermicro/results/SYS-822GS-NBRT_B200-SXM-180GB_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0177, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "NVIDIA GB200 NVL72 by HPE (4x NVIDIA GB200)", "platform_url": "/platform/nvidia-gb200-nvl72-by-hpe-4x-nvidia-gb200.html", "perf": 101453, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, TensorRT-LLM 1.3.0rc0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "HPE", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system HPE_GB200-NVL72_GB200-186GB_aarch64x8", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/HPE/results/HPE_GB200-NVL72_GB200-186GB_aarch64x8", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0176, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "QuantaGrid D74A-7U (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/quantagrid-d74a-7u-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 100754, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TODO", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Supermicro", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0097", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Supermicro/results/8xMI355X_2xEPYC-9575F", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0175, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Dell PowerEdge XE9785L (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/dell-poweredge-xe9785l-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 100008, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "vLLM 0.14.0rc3.dev30+gb026cf14e.d20260212.rocm700, Pytorch 2.9.1+git8907517, ROCm 7.0.0.70000-38~22.04 ", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0020", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Dell/results/8xMI355X_2xEPYC_9655", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0174, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "G893-ZX1-AAX4 (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/g893-zx1-aax4-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 100310, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "PyTorch 2.9.1+git8907517, ROCm 7.0.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "GigaComputing", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0037", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/GigaComputing/results/8xMI355X_2xEPYC-9575F", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0174, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "BM.GPU.MI355X.8 (8x AMD Instinct MI355X 288GB HBM3e)", "platform_url": "/platform/bm-gpu-mi355x-8-8x-amd-instinct-mi355x-288gb-hbm3e.html", "perf": 99874.3, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "ORACLE", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0086", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/ORACLE/results/8xMI355X_2xEPYC_9575F", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0174, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Inventec P9000AG7 (8x NVIDIA B200-SXM-180GB)", "platform_url": "/platform/inventec-p9000ag7-8x-nvidia-b200-sxm-180gb.html", "perf": 100291, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.11, CUDA 12.9", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "B", "confidence_label": "B – documented third-party method", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Inventec", "source": { "name": "MLPerf Inference v6.0 (open, datacenter) ID 6.0-0105", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/open/Inventec/results/P9000AG7_B200-SXM-180GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0174, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Cisco UCS H200 x8 + MI350X x8 G200 2x2 TOR (8x NVIDIA H200-SXM-141GB + AMD Instinct MI350X 288GB HBM3e)", "platform_url": "/platform/cisco-ucs-h200-x8-mi350x-x8-g200-2x2-tor-8x-nvidia-h200-sxm-141gb-amd-instinct-m.html", "perf": 99683.8, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp8,fp4", "stack": "vLLM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Cisco", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system Cisco_UCS_H200x8_MI350Xx8_G200_2x2_TOR", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Cisco/results/Cisco_UCS_H200x8_MI350Xx8_G200_2x2_TOR", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0173, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "G8825Z5 (8x AMD Instinct MI350X 288GB HBM3e)", "platform_url": "/platform/g8825z5-8x-amd-instinct-mi350x-288gb-hbm3e.html", "perf": 97019.7, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "Pytorch 2.10, ROCm 7.2.2", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "MiTAC", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system 8xMI350X_2xEPYC-9755", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/MiTAC/results/8xMI350X_2xEPYC-9755", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0169, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "LLM-D v0.5.0,Openshift 4.20.12,NVIDIA 8xB200-SXM-180GB (8x NVIDIA B200-SXM-180GB)", "platform_url": "/platform/llm-d-v0-5-0-openshift-4-20-12-nvidia-8xb200-sxm-180gb-8x-nvidia-b200-sxm-180gb.html", "perf": 93070.7, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "mxfp4", "stack": "LLM-D v0.5.0 ,vLLM 0.14.1, RHEL 9.6", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "RedHat", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0091", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/RedHat/results/8xB200-LLM-D-Openshift", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0162, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Nebius B200 n1 (8x NVIDIA B200-SXM-180GB)", "platform_url": "/platform/nebius-b200-n1-8x-nvidia-b200-sxm-180gb.html", "perf": 89856.3, "perf_unit": "tok/s", "metric": "throughput_server", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Nebius", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system B200-SXM-180GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Nebius/results/B200-SXM-180GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0156, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "PowerEdge XE9785 (8x AMD Instinct MI355X 288GB HBM3e (Power Cap 1000 W))", "platform_url": "/platform/poweredge-xe9785-8x-amd-instinct-mi355x-288gb-hbm3e-power-cap-1000-w.html", "perf": 81870.5, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp8", "stack": "ROCm", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell_MangoBoost", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0033", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Dell_MangoBoost/results/8xMI355X_2xEPYC_9965_Power_Cap_1000W", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0142, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "AS-8126GS-TNMR,vLLM (8x AMD Instinct MI350X 288GB HBM3e)", "platform_url": "/platform/as-8126gs-tnmr-vllm-8x-amd-instinct-mi350x-288gb-hbm3e.html", "perf": 80478.1, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "vLLM 0.14.0, RHEL 9.6", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Supermicro_RedHat", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0100", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Supermicro_RedHat/results/8XMI350x_RHEL9", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.014, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Cisco UCS 8xMI350X (8x AMD Instinct MI350X 288GB HBM3e)", "platform_url": "/platform/cisco-ucs-8xmi350x-8x-amd-instinct-mi350x-288gb-hbm3e.html", "perf": 78936.7, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "vLLM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Cisco", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0012", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Cisco/results/Cisco_UCS_8xMI350x", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0137, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Lambda 4x GB300-288GB (4x NVIDIA GB300)", "platform_url": "/platform/lambda-4x-gb300-288gb-4x-nvidia-gb300.html", "perf": 65511.9, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "mxfp4", "stack": "TensorRT-LLM 1.3.0rc14 (GPT-OSS-120B); vLLM (CentML) 0.20.2rc1.dev2091+g2e1040b7e + NVIDIA Dynamo (Qwen3-VL-235B); CUDA ", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Lambda", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system GB300-NVL4_GB300-288GB_aarch64x4", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Lambda/results/GB300-NVL4_GB300-288GB_aarch64x4", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0114, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "AS-8126GS-TNMR,vLLM (8x AMD Instinct MI350X)", "platform_url": "/platform/as-8126gs-tnmr-vllm-8x-amd-instinct-mi350x.html", "perf": 64293.3, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "B", "confidence_label": "B – documented third-party method", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Supermicro_RedHat", "source": { "name": "MLPerf Inference v6.0 (open, datacenter) ID 6.0-0112", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/open/Supermicro_RedHat/results/8xMI350X_2xEPYC-9575F", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0112, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "OneW GB300 NVL72 (4x NVIDIA GB300)", "platform_url": "/platform/onew-gb300-nvl72-4x-nvidia-gb300.html", "perf": 63990.5, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Wiwynn", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system GB300-NVL72_GB300-288GB_aarch64x4_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Wiwynn/results/GB300-NVL72_GB300-288GB_aarch64x4_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0111, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "SMC Lab OpenShift GB200-NVL4 4xGB200-186GB (4x NVIDIA GB200)", "platform_url": "/platform/smc-lab-openshift-gb200-nvl4-4xgb200-186gb-4x-nvidia-gb200.html", "perf": 61205.6, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp8", "stack": "vLLM 0.24.0, Red Hat OpenShift 4.21", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "RedHat_Supermicro", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system GB200-NVL4_186GB_Openshift_VLLM", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/RedHat_Supermicro/results/GB200-NVL4_186GB_Openshift_VLLM", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0106, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "XE7745_8xMI350P_2xEPYC_9455 (8x AMD Instinct MI350P)", "platform_url": "/platform/xe7745-8xmi350p-2xepyc-9455-8x-amd-instinct-mi350p.html", "perf": 48971.1, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "vLLM 0.22.0, PyTorch 2.10.0, ROCm 7.2.2", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell_AMD", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system XE7745_8xMI350P_2xEPYC-9455", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Dell_AMD/results/XE7745_8xMI350P_2xEPYC-9455", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0085, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "HPE ProLiant Compute DL380a Gen12 (10x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/hpe-proliant-compute-dl380a-gen12-10x-nvidia-rtx-pro-6000-blackwell-server-editi.html", "perf": 36710.8, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.14.1.48, CUDA 13.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "HPE", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0045", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/HPE/results/HPE_PROLIANT_DL380A_RTX_PRO_6000_PCIE_96GBx10_TRT_10_14_1_48_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0064, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "MangoBoost Mi325X (8x AMD Instinct MI325X 256GB HBM3e)", "platform_url": "/platform/mangoboost-mi325x-8x-amd-instinct-mi325x-256gb-hbm3e.html", "perf": 33880.4, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp8", "stack": "ROCm 6.10.5", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "MangoBoost", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0070", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/MangoBoost/results/8xMI325X_2xEPYC_9655", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0059, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Dell PowerEdge XE7740 (8x NVIDIA H200-NVL-141GB)", "platform_url": "/platform/dell-poweredge-xe7740-8x-nvidia-h200-nvl-141gb.html", "perf": 32004, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp8", "stack": "TensorRT 10.13, CUDA 13.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0021", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Dell/results/XE7740_H200_NVL_141GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0056, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "8xMI300X_2xEPYC_9534 (8x AMD Instinct MI300X 192GB HBM3)", "platform_url": "/platform/8xmi300x-2xepyc-9534-8x-amd-instinct-mi300x-192gb-hbm3.html", "perf": 30198.4, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp8", "stack": "vLLM 0.24.0 (stock vllm/vllm-openai-rocm:v0.24.0), ROCm 7.2.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "MangoBoost", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system 8xMI300X_2xEPYC_9534", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/MangoBoost/results/8xMI300X_2xEPYC_9534", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0053, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Cisco UCS C845A M8 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/cisco-ucs-c845a-m8-8x-nvidia-rtx-pro-6000-blackwell-server-edition.html", "perf": 29904, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Cisco", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system C845A_M8_RTXPro6000_96GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Cisco/results/C845A_M8_RTXPro6000_96GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0052, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Dell PowerEdge XE7740 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/dell-poweredge-xe7740-8x-nvidia-rtx-pro-6000-blackwell-server-edition.html", "perf": 29796.7, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system XE7740_RTXPro6000_PCIe_96GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Dell/results/XE7740_RTXPro6000_PCIe_96GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0052, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "HPE ProLiant Compute DL380a Gen12 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/hpe-proliant-compute-dl380a-gen12-8x-nvidia-rtx-pro-6000-blackwell-server-editio.html", "perf": 29908.4, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.14.1.48, CUDA 13.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "HPE", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0047", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/HPE/results/HPE_PROLIANT_DL380A_RTX_PRO_6000_PCIE_96GBx8_TRT_10_14_1_48_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0052, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "MiTAC_G4520G6_RTX_PRO_6000_PCIE_96GBx8 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/mitac-g4520g6-rtx-pro-6000-pcie-96gbx8-8x-nvidia-rtx-pro-6000-blackwell-server-e.html", "perf": 30034.2, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.2", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "MiTAC", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system MiTAC_G4520G6_RTX_PRO_6000_PCIE_96GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/MiTAC/results/MiTAC_G4520G6_RTX_PRO_6000_PCIE_96GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0052, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "LLM-D v0.5.0,OpenShift 4.21,IBM cloud gx3d-160x1792x8h200,NVIDIA 8xH200-SXM-141GB (8x NVIDIA H200-SXM-141GB)", "platform_url": "/platform/llm-d-v0-5-0-openshift-4-21-ibm-cloud-gx3d-160x1792x8h200-nvidia-8xh200-sxm-141g.html", "perf": 28680, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "mxfp4", "stack": "LLM-D v0.5.0 ,OpenShift 4.21, RHEL 9.6", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "RedHat", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0092", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/RedHat/results/8xH200-LLM-D-Openshift", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.005, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "ESC8000A_E13P_RTX_PRO_6000_PCIE_96GBx8 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/esc8000a-e13p-rtx-pro-6000-pcie-96gbx8-8x-nvidia-rtx-pro-6000-blackwell-server-e.html", "perf": 28038.5, "perf_unit": "tok/s", "metric": "throughput_server", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "ASUSTeK", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0004", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/ASUSTeK/results/ESC8000A_E13P_RTX_PRO_6000_PCIE_96GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0049, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "PRIMERGY_CDI_RTX_PRO_6000_PCIE_96GBx8) (8x NVIDIA RTX PRO 6000 Blackwell)", "platform_url": "/platform/primergy-cdi-rtx-pro-6000-pcie-96gbx8-8x-nvidia-rtx-pro-6000-blackwell.html", "perf": 27340.9, "perf_unit": "tok/s", "metric": "throughput_server", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.16, CUDA 13.2, cuDNN 9.21", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Fujitsu", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system PRIMERGY_CDI_RTX_PRO_6000_PCIE_96GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Fujitsu/results/PRIMERGY_CDI_RTX_PRO_6000_PCIE_96GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0048, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "ESC8000_E12P_RTX_PRO_6000_PCIE_96GBx8 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/esc8000-e12p-rtx-pro-6000-pcie-96gbx8-8x-nvidia-rtx-pro-6000-blackwell-server-ed.html", "perf": 27730.1, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "ASUSTeK", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0005", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/ASUSTeK/results/ESC8000_E12P_RTX_PRO_6000_PCIE_96GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0048, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "MangoBoost Mi300X (8x AMD Instinct MI300X 192GB HBM3)", "platform_url": "/platform/mangoboost-mi300x-8x-amd-instinct-mi300x-192gb-hbm3.html", "perf": 27598.7, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp8", "stack": "ROCm 6.10.5", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "MangoBoost", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0069", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/MangoBoost/results/8xMI300X_2xEPYC_9534", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0048, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Dell PowerEdge XE9680 (8x AMD Instinct MI300X 192GB HBM3)", "platform_url": "/platform/dell-poweredge-xe9680-8x-amd-instinct-mi300x-192gb-hbm3.html", "perf": 27072.5, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp8", "stack": "vLLM 0.9.0.2.dev108+g71faa1880.d20260213, Pytorch 2.7.0+gitf717b2a, ROCm 6.4.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0019", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Dell/results/8xMI300X_2xPlatinum_8470", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0047, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Dell PowerEdge XE7745 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/dell-poweredge-xe7745-8x-nvidia-rtx-pro-6000-blackwell-server-edition.html", "perf": 26829.8, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0023", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Dell/results/XE7745_RTXPro6000_PCIe_96GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0047, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Lenovo ThinkSystem SR675i V3 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/lenovo-thinksystem-sr675i-v3-8x-nvidia-rtx-pro-6000-blackwell-server-edition.html", "perf": 27160.2, "perf_unit": "tok/s", "metric": "throughput_server", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Lenovo", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0065", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Lenovo/results/SR675iV3", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0047, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "HPE Cray XD670 (8x NVIDIA H200-SXM-141GB)", "platform_url": "/platform/hpe-cray-xd670-8x-nvidia-h200-sxm-141gb.html", "perf": 25939.8, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp16", "stack": "vLLM v0.15.0-dev", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "B", "confidence_label": "B – documented third-party method", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Krai", "source": { "name": "MLPerf Inference v6.0 (open, datacenter) ID 6.0-0109", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/open/Krai/results/h200_n1-vllm", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0045, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "NVIDIA DGX B200 (8x NVIDIA B200-SXM-180GB)", "platform_url": "/platform/nvidia-dgx-b200-8x-nvidia-b200-sxm-180gb.html", "perf": 16167.4, "perf_unit": "tok/s", "metric": "throughput_server", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "B", "confidence_label": "B – documented third-party method", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Lambda_SIT", "source": { "name": "MLPerf Inference v6.0 (open, datacenter) ID 6.0-0110", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/open/Lambda_SIT/results/B200-SXM-180GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0028, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Nebius RTX PRO 6000 n1 (8x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/nebius-rtx-pro-6000-n1-8x-nvidia-rtx-pro-6000-blackwell-server-edition.html", "perf": 15738.5, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Nebius", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system RTX_PRO_6000_PCIE_96GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Nebius/results/RTX_PRO_6000_PCIE_96GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0027, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Nebius GB300 NVL72 (1x NVIDIA GB300)", "platform_url": "/platform/nebius-gb300-nvl72-1x-nvidia-gb300.html", "perf": 15371.7, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf gpt-oss-120b", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, Driver 580.105.08, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Nebius", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0079", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Nebius/results/GB300-NVL72_GB300-288GB_aarch64x1_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0027, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Cisco UCS X210c M8 - X580P PCIe Node (4x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/cisco-ucs-x210c-m8-x580p-pcie-node-4x-nvidia-rtx-pro-6000-blackwell-server-editi.html", "perf": 14860.7, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Cisco", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system X210M8_X580P_RTXPro6000_96GBx4_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Cisco/results/X210M8_X580P_RTXPro6000_96GBx4_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0026, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "B300-SXM-270GBx1_TRT (1x NVIDIA B300-SXM-270GB)", "platform_url": "/platform/b300-sxm-270gbx1-trt-1x-nvidia-b300-sxm-270gb.html", "perf": 14636.8, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "VibeHPC", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system B300-SXM-270GBx1_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/VibeHPC/results/B300-SXM-270GBx1_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0025, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Cisco UCS X215c M8 - X580P PCIe Node (4x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/cisco-ucs-x215c-m8-x580p-pcie-node-4x-nvidia-rtx-pro-6000-blackwell-server-editi.html", "perf": 13529.8, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.11, CUDA 12.9", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Cisco", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0014", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Cisco/results/X215M8_X580P_RTXPro6000_96GBx4_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0024, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "HPE ProLiant DL385 Gen11 (4x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/hpe-proliant-dl385-gen11-4x-nvidia-rtx-pro-6000-blackwell-server-edition.html", "perf": 13361.9, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.14.1.48, CUDA 13.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "HPE", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0054", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/HPE/results/HPE_Proliant_DL385_RTX_PRO_6000_PCIe_96GBx4_TRT_10_14_1_48_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0023, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "HPE ProLiant DL385 Gen11 (2x NVIDIA RTX PRO 6000 Blackwell Server Edition)", "platform_url": "/platform/hpe-proliant-dl385-gen11-2x-nvidia-rtx-pro-6000-blackwell-server-edition.html", "perf": 6862.3, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "fp4", "stack": "TensorRT 10.11.0.33, CUDA 13.0", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "HPE", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0051", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/HPE/results/HPE_Proliant_DL385_RTX_PRO_6000_PCIe_96GBx2_TRT_10_11_0_33_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0012, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "Dell PowerEdge XE7745 (8x NVIDIA RTX PRO 4500 Blackwell)", "platform_url": "/platform/dell-poweredge-xe7745-8x-nvidia-rtx-pro-4500-blackwell.html", "perf": 4374.06, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "fp4", "stack": "TensorRT 10.14, CUDA 13.3", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Dell", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system XE7745_RTXPRO4500_32GBx8_TRT", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Dell/results/XE7745_RTXPRO4500_32GBx8_TRT", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0008, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "1-node-4x-BMG-Pro-B60-Dual (4x MS-Intel Arc Pro B60 Dual 48G Turbo)", "platform_url": "/platform/1-node-4x-bmg-pro-b60-dual-4x-ms-intel-arc-pro-b60-dual-48g-turbo.html", "perf": 3270.66, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "INT8", "stack": "PyTorch vLLM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Intel", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0060", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Intel/results/1-node-4x-BMG-Pro-B60-Dual", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0006, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "1-node-4x-BMG-B70 (4x Intel(R) Arc(TM) Pro B70)", "platform_url": "/platform/1-node-4x-bmg-b70-4x-intel-r-arc-tm-pro-b70.html", "perf": 2464.98, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99.9", "precision": "INT8", "stack": "PyTorch", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-09-16 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Intel", "source": { "name": "MLPerf Inference v6.1 (closed, datacenter) system 1-node-4x-BMG-B70", "url": "https://github.com/mlcommons/inference_results_v6.1/tree/main/closed/Intel/results/1-node-4x-BMG-B70", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0004, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "1-node-4x-BMG-B70 (4x Intel(R) Arc Pro(R) B70)", "platform_url": "/platform/1-node-4x-bmg-b70-4x-intel-r-arc-pro-r-b70.html", "perf": 2459.18, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "INT8", "stack": "PyTorch vLLM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Intel", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0101", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Intel/results/1-node-4x-BMG-B70", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0004, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 }, { "platform": "1-node-4x-BMG-B60 (4x Intel(R) Arc Pro(R) B60)", "platform_url": "/platform/1-node-4x-bmg-b60-4x-intel-r-arc-pro-r-b60.html", "perf": 1697.66, "perf_unit": "tok/s", "metric": "throughput_offline", "model": "MLPerf llama2-70b-99", "precision": "INT8", "stack": "PyTorch vLLM", "size_class": "LLM >40B", "batch": "unknown", "like_for_like": true, "differs": [], "confidence": "A", "confidence_label": "A – audited (MLPerf closed)", "power_w": null, "power_label": "no power published", "power_kind": "unknown", "perf_per_w": null, "date": "2026-04-01 (published)", "older_than_18_months": false, "superseded": false, "power_basis": "unknown", "price_usd": null, "price_label": "unknown", "price_date": null, "price_launch_usd": null, "edge": false, "date_basis": "published", "form_factor": "datacenter server", "vendor": "Intel", "source": { "name": "MLPerf Inference v6.0 (closed, datacenter) ID 6.0-0059", "url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/Intel/results/1-node-4x-BMG-B60", "date": null }, "power_source": null, "price_source": null, "fit": { "perf": null, "power": null, "budget": null }, "meets_all": true, "unknowns": 0, "eligible": true, "headroom": null, "rank_score": 0.0003, "unknown_badges": [ "power unknown", "perf/W unknown", "price unknown" ], "evidence_tier": 1 } ], "no_data": null, "legend": "Confidence: A audited MLPerf closed, B documented third-party, C community, D vendor-published, E estimate. Unknown is never assumed." }