NVIDIA · NVIDIA GB200 NVL72 (72x GB200-186GB_aarch64, TensorRT)
Llama-2-70B (MLPerf reference) · llama2-70b-99 · Offline · run 01a00b4a-2a72…
valid · Reported evidence
Measurements
output_token_throughput_tps · reported888,054 tokens_s
Llama-2-70B (MLPerf reference) · llama2-70b-99 · Offline · offline · 4× NVIDIA GB200 × 18 nodes · mlperf_closed_99_9 · observed 2026-03-26
Identity
modelLlama-2-70B (MLPerf reference)
stackTensorRT 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
workloadllama2-70b-99 · Offline
topology4× NVIDIA GB200 × 18 nodes
harnessmlperf_loadgen v6.0
quality policymlperf_closed_99_9
comparison keysha256:1e3cbb87da553569…
sha256:7d4a26897cca44becfc416f…
Launch configuration
dtypefp4
notesNVIDIA GB200 NVL72, 1200W TGP Per GPU
systemNVIDIA GB200 NVL72 (72x GB200-186GB_aarch64, TensorRT)
softwareTensorRT 10.14, CUDA 13.1, cuDNN 9.17, TensorRT-LLM feat/1.2-mlpinf, NVIDIA Dynamo mlperf-v6.0-dynamo-v0.8.0, vLLM CentML:mlperf-inf-mm-q3vl-v6.0
availabilityavailable
Workload
namellama2-70b-99 · Offline
streamingno
load generationoffline
Canonical manifest
Show manifestHide manifest
{
"kind": "ServingRun",
"spec": {
"status": "valid",
"harness": {
"name": "mlperf_loadgen",
"version": "v6.0"
},
"topology": {
"nodeCount": 18,
"acceleratorModel": "NVIDIA GB200",
"acceleratorVendor": "nvidia",
"acceleratorsPerNode": 4
},
"observedAt": "2026-03-26T19:39:54Z",
"modelDigest": "sha256:8005d6af1339caa02e60709ce2d7a6002ce9c51e436d21bb39dc5b8f93a24b0c",
"measurements": [
{
"unit": "tokens_s",
"value": 888054,
"metric": "output_token_throughput_tps",
"statistic": "reported"
}
],
"sourceNative": {
"source": "mlperf-inference",
"metrics": {
"performance_units": "Tokens/s",
"performance_result": 888054
},
"category": "datacenter",
"division": "closed",
"scenario": "offline",
"benchmark": "llama2-70b-99.9",
"externalId": "v6.0/6.0-0075"
},
"qualityPolicy": "mlperf_closed_99_9",
"workloadDigest": "sha256:65632b1fcd025d277521ab3869f904644b69e0e5322392b1d05ab8a0be7d84b9",
"configurationDigest": "sha256:65fb87c0e4d4c4a1f32024f738291a53d1e243bac2f704a29cdf7a9d9bd977bc"
},
"metadata": {
"name": "mlperf-v6-0-6-0-0075",
"title": "NVIDIA · llama2-70b-99.9 · Offline",
"sourceRefs": [
{
"url": "https://github.com/mlcommons/submissions_inference_v6.0/tree/main/closed/NVIDIA/results/GB200-NVL72_GB200-186GB_aarch64x72_TRT"
}
]
},
"apiVersion": "kernelindex.dev/v1alpha1"
}Cite this result (permalink, digest, access date)
Report an issue with this serving run
MLPerf Inference · closed/datacenter · entry v6.0/6.0-0075 · shown unmodified as publishedAll serving results →JSON