Ben Linville (bentech) - Intel Arc Pro B60 + B70 LLM benchmarks, most quants every backend

25-run matrix Apr 16 2026 on Arc Pro B70 32GB + B60 24GB: llama-benchy 0.3.5, llama.cpp b1-ff5ef82 Vulkan+SYCL, Intel llm-scaler-vllm:0.14.0-b8.1 Docker. pp512/tg128, 3 runs/test, q8_0 KV, FA on, 8k ctx. Qwen3.5-27B all quants + Gemma-4-31B. Dual-GPU over PCIe4 x4 M.2 slower than single.