MLPerf v6.1 NVIDIA DGX B300 (8x B300-SXM-270GB) llama2-70b-99.9 Server run_1 harness summary log

Fields

id

sources/bench-mlperf-v6-1-nvidia-b300-llama2-server-log

URL

https://raw.githubusercontent.com/mlcommons/inference_results_v6.1/main/closed/NVIDIA/results/B300-SXM-270GBx8_TRT/llama2-70b-99.9/Server/performance/run_1/mlperf_log_summary.txt

Publisher

MLCommons

Kind

benchmark

Accessed

2026-10-04

notes

Result is : VALID. Completed samples per second 391.79, Completed tokens per second 106949.36. Crucially this log carries a line the Offline log does not: 'Performance constraints satisfied : Yes', plus explicit 'TTFT Early Stopping Result: Run successful' and 'TPOT Early Stopping Result: Run successful'. The 2.2% Offline-over-Server gap on this node is the price of the Server scenario's per-query TTFT/TPOT bound and its target_qps=0.5 Poisson arrival rate.

Notes

Result is : VALID. Completed samples per second 391.79, Completed tokens per second 106949.36. Crucially this log carries a line the Offline log does not: 'Performance constraints satisfied : Yes', plus explicit 'TTFT Early Stopping Result: Run successful' and 'TPOT Early Stopping Result: Run successful'. The 2.2% Offline-over-Server gap on this node is the price of the Server scenario's per-query TTFT/TPOT bound and its target_qps=0.5 Poisson arrival rate.

Cross-references

This record references no other records.