inference/research
verifiedmedium · 0.75updated 2026-10-15

Argonne GH200 x1 TRT-LLM Mistral-7B b64

15306.2 tok/s aggregate

Fields

id
benchmarks/argonne-gh200-trtllm-mistral-7b-fp16-b64-128io-tps
Metric
tps_aggregate
Value
15,306.2
Unit
tok/s aggregate
Model
mistralai/Mistral-7B-v0.1
Methodology
THIRD-PARTY MEASURED (Argonne LLM-Inference-Bench arXiv:2411.00136). tensorrt-llm, FP16/BF16, 128in/128out. throughput=batch*(in+out)/e2e latency = AGGREGATE across batch. CSV 'Nvidia GH200 GPU' = GH200 Grace Hopper superchip (H100 GPU + Grace CPU over NVLink-C2C). Single-GPU rows only; the CSV has no multi-GPU GH200. bs64 lat 1.07s.
Measured by
third_party
Reproducible
yes
notes
GH200 superchip; on-package Grace CPU.

Notes

GH200 superchip; on-package Grace CPU.

Cross-references