Level1Techs: 4x Intel Arc Pro B70, vLLM, Qwen3.5-27B, 50 concurrent: avg TTFT ~11.4 s
11400.0 ms mean time-to-first-token under 50-request load
Fields
id
benchmarks/l1t-4x-arc-pro-b70-qwen35-27b-vllm-50req-ttftMetric
ttft_ms
Value
11,400
Unit
ms mean time-to-first-token under 50-request load
Model
Qwen3.5-27B
Methodology
FORUM MEASUREMENT (Level1Techs via hardware-corner). 4x B70 vLLM Qwen3.5-27B 50 concurrent ~1024 tok: avg TTFT ~11.4 s (vs ~18.7 s on 4x RTX 3090 in the same test).
Measured by
third_party
Reproducible
no
notes
TTFT under saturation, not single-request.
Notes
TTFT under saturation, not single-request.