Level1Techs: 4x Intel Arc Pro B70, vLLM, Qwen3.5-27B, 50 concurrent reqs ~1024 tok ctx: ~369 tok/s aggregate output
369.0 tok/s output, aggregate over 50 concurrent requests
Fields
id
benchmarks/l1t-4x-arc-pro-b70-qwen35-27b-vllm-50req-tpsMetric
tps_aggregate
Value
369
Unit
tok/s output, aggregate over 50 concurrent requests
Model
Qwen3.5-27B
Methodology
FORUM MEASUREMENT (Level1Techs, relayed by hardware-corner.net B70 launch article). 4x Arc Pro B70 vLLM, Qwen3.5-27B, 50 concurrent requests, ~1024-token context: sustained ~369 tok/s output (peak ~550), avg TTFT ~11.4 s. AGGREGATE across 50 requests, not per-request.
Measured by
third_party
Reproducible
no
notes
Companion record l1t-4x-rtx-3090-qwen35-27b-vllm-50req-tps gives the same-test 4x3090 baseline (348 tok/s, TTFT 18.7s).
Notes
Companion record l1t-4x-rtx-3090-qwen35-27b-vllm-50req-tps gives the same-test 4x3090 baseline (348 tok/s, TTFT 18.7s).