Intel Arc Pro B70 32GB x4, MLPerf v6.0 Offline: gpt-oss-120b aggregate tok/s
1536.9 output tok/s AGGREGATE across 4x Arc Pro B70 (Offline; ~384 tok/s per card)
Fields
id
benchmarks/b70-4x-mlperf-v6-0-gptoss-120b-offline-tokensMetric
tps_aggregate
Value
1,536.9
Unit
output tok/s AGGREGATE across 4x Arc Pro B70 (Offline; ~384 tok/s per card)
Model
gpt-oss-120b (FP8)
Methodology
VENDOR-MEASURED (MLPerf v6.0 closed submission, Intel). 4x Arc Pro B70 32GB GDDR6 (PCIe Gen5 x16) on Xeon 698X host, PyTorch vLLM, gpt-oss-120b, Offline. 'Tokens per second' = 1536.9 verbatim from mlperf_log_summary.txt. AGGREGATE across 4 cards — ~384/card.
Measured by
vendor
Reproducible
yes
notes
gpt-oss-120b MoE on 4x B70 — ~384 tok/s per card aggregate.
Notes
gpt-oss-120b MoE on 4x B70 — ~384 tok/s per card aggregate.