Intel Arc Pro B50 x4, MLPerf v6.0 Server: Llama-3.1-8B INT8 aggregate tok/s
1187.13 output tok/s AGGREGATE across 4x Arc Pro B50 (Server scenario; latency-constrained)
Fields
id
benchmarks/b50-4x-mlperf-v6-0-llama31-8b-server-tokensMetric
tps_aggregate
Value
1,187.13
Unit
output tok/s AGGREGATE across 4x Arc Pro B50 (Server scenario; latency-constrained)
Model
Llama-3.1-8B (INT8)
Methodology
VENDOR-MEASURED (MLPerf Inference v6.0 closed submission, Intel). System 1-node-4x-BMG-B50: 4x [MAXSUN] MS-Intel Arc Pro B50 16G (Battlemage, 16 GB GDDR6, 224 GB/s, PCIe Gen5 x8 electrical) on one Intel Xeon Platinum 8480C host, Ubuntu 25.04, PyTorch vLLM, Llama-3.1-8B INT8 weights. Server scenario. 'Tokens per second' = 1187.13 verbatim from the submission summary (run_1). AGGREGATE across all 4 accelerators under the MLPerf Server latency constraint, NOT per-request and NOT a batch-1 decode rate. Accuracy log reports GEN_LEN 8149111 (ROUGE1 38.7283).
Measured by
vendor
Reproducible
yes
notes
Lower than the Offline number on the same system because the Server scenario is latency-bounded. Do not read 1187.13 as a single-stream tok/s.
Notes
Lower than the Offline number on the same system because the Server scenario is latency-bounded. Do not read 1187.13 as a single-stream tok/s.