Intel Arc B580/B70 ComfyUI Benchmarks: omnixpu Beats Native pytorchxpu
Willing-Grade5566 · reddit · 2026-10-31
The author runs Intel's official llm-scaler/omni/ Omni acceleration in Windows ComfyUI, compiling smoothly on both Arc B580 and B70 with noticeable gains, and benchmarks omnixpu against native pytorchxpu (logs attached).
- Official repo targets BMG (Battlemage) GPUs—B580/B60/B65/B70—with omnixpukernel, ComfyUI-OmniXPU plugin, plus community aimdo/kitchen xpu providers.
- Community DG2 (Alchemist, e.g. A770) builds exist, bypassing kitchenxpu and injecting operator routing directly into the omnixpu plugin.
More from Infra
- Lawyer runs DeepSeek-V4.1-Flash on 2x DGX Spark + 3090s: 2,618 tok/s prefill at 256K, 95.9% KLD accuracy, all data open — TumbleweedNew6515 · 2026-10-31
- Koboldcpp v1.124 "aesthetic edition" released for local LLM inference — Fcking_Chuck · 2026-10-31
- CXMT validates G6 DRAM engineering samples in Q3, mass production expected within a year — zephyr_z9 · 2026-10-31
- AI scarcity moved to atoms: memory, GPU and wafer margins hit 58%-80% — sujingshen · 2026-10-31
- Anthropic's S-1 risk factors: compute tops the list at 13.4%, edging out regulation — ayushtweetshere · 2026-10-31
- TU Eindhoven shows glass interposer with BGA below and MPO optical port on top — jwt0625 · 2026-10-31