vLLM 0.12 on 4x Arc Pro B60 (TP=4): Qwen3-VL-30B peak aggregate throughput (c=64)
1000 tok/s aggregate (peak, c=64)
Fields
id
benchmarks/embeddedllm-b60x4-vllm-qwen3vl30b-peak-tpsMetric
tps_aggregate
Value
1,000
Unit
tok/s aggregate (peak, c=64)
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. ~1000 tok/s aggregate at 64 concurrent requests (long-form generation scenario).
Measured by
third_party
Reproducible
yes
notes
Peak aggregate across 64 concurrent requests - server throughput, not per-request decode.
Notes
Peak aggregate across 64 concurrent requests - server throughput, not per-request decode.