vLLM 0.12 on 4x Arc Pro B60 (TP=4): Qwen3-VL-30B chat 1K->1K TPOT (concurrency 16)
61 ms/token
Fields
id
benchmarks/embeddedllm-b60x4-vllm-qwen3vl30b-chat-tpotMetric
itl_ms
Value
61
Unit
ms/token
Model
Qwen/Qwen3-VL-30B-A3B-Instruct
Methodology
THIRD-PARTY MEASURED (Embedded LLM, Feb 14 2026). Sparkle 4x Intel Arc Pro B60 workstation, TP=4, 24GB/456GB-s per GPU. Qwen/Qwen3-VL-30B-A3B-Instruct. vLLM 0.12 vs Intel LLM-Scaler 1.2. Concurrency 16/32/64, 1K->1K and 1K->2K scenarios. Chat scenario 1K->1K at concurrency 16. TPOT = inter-token latency (per-request decode).
Measured by
third_party
Reproducible
yes
notes
vLLM 61ms/token vs LLM-Scaler 48ms/token (Intel build ~25% faster decode streaming). ~1000 tok/s aggregate peak at c=64.
Notes
vLLM 61ms/token vs LLM-Scaler 48ms/token (Intel build ~25% faster decode streaming). ~1000 tok/s aggregate peak at c=64.