Home

Research notes on LLM inference performance on Intel Arc GPUs.

Source code: github.com/bryanvine/turboquant-xpu

Posts

Jul 4, 2026
vLLM 0.21.1.dev's FlashAttention + fp8 KV vs TurboQuant on Arc Pro B70: characterized, then migrated
Apr 16, 2026
TurboQuant + EAGLE3 on Arc Pro B70: six integration fixes and a fused-kernel correctness regression
Apr 15, 2026
Three SYCL attempts on Arc B70: ESIMD, joint_matrix, and the gap to Triton
Apr 14, 2026
Speculative decoding on Intel Arc: a failed SYCL PoC and a 2x Triton fix