NVIDIA H100 NVL 94GB x8 (Lenovo SR675 V3), MLPerf Offline (v4.1): GPT-J 6B aggregate tok/s
14073.6 output tok/s AGGREGATE across 8x H100-NVL-94GB (Offline (v4.1); ~1759 tok/s per card)
Fields
id
benchmarks/h100nvl-8x-mlperf-v4-1-gptj-offline-tokensMetric
tps_aggregate
Value
14,073.6
Unit
output tok/s AGGREGATE across 8x H100-NVL-94GB (Offline (v4.1); ~1759 tok/s per card)
Model
GPT-J 6B (FP8)
Methodology
VENDOR-MEASURED (MLPerf closed submission, Lenovo SR675 V3). 8x H100-NVL-94GB (PCIe, 3-card NVLink bridge pair), TensorRT, GPT-J 6B FP8, Offline (v4.1). 'Tokens per second' = 14073.6 verbatim from mlperf_log_summary.txt. AGGREGATE across 8 cards — NOT a single-user decode figure (~1759/card).
Measured by
vendor
Reproducible
yes