Intel Arc Pro B60 24GB x4, MLPerf v5.1 Offline: Llama-2-70B aggregate tok/s
3009.38 output tok/s AGGREGATE across 4x Arc Pro B60 (Offline; ~752 tok/s per card)
Fields
id
benchmarks/prob60dual-4x-mlperf-v5-1-llama2-70b-offline-tokensMetric
tps_aggregate
Value
3,009.38
Unit
output tok/s AGGREGATE across 4x Arc Pro B60 (Offline; ~752 tok/s per card)
Model
Llama-2-70B (FP8)
Methodology
VENDOR-MEASURED (MLPerf v5.1 closed submission, Intel). 4x Arc Pro B60 24GB GDDR6 (PCIe Gen5 x8) on Xeon 698X host, PyTorch vLLM, Llama-2-70B, Offline. 'Tokens per second' = 3009.38 verbatim from mlperf_log_summary.txt. AGGREGATE across 4 cards — ~752/card.
Measured by
vendor
Reproducible
yes
notes
4x 'BMG-Pro-B60-Dual' (dual-GPU card = 8 B60 dies) — per-DIE ~376 tok/s.
Notes
4x 'BMG-Pro-B60-Dual' (dual-GPU card = 8 B60 dies) — per-DIE ~376 tok/s.