Intel 285HX Tests Show CPU Beating iGPU for Local LLMs

A user benchmarked local LLM inference on an Intel 285HX mini PC across llama.cpp backends and models, comparing CPU, Vulkan, and SYCL. The takeaway was that Intel’s Arrow Lake iGPU performed poorly, with CPU often delivering the better result for local runs.

2026-07-12 ~ 2026-07-13 · 2 related posts

• Running Local LLMs on 285HX: CPU is Surprisingly the Fastest — nirurin · 2026-07-12
• llama.cpp Underperforms on Intel iGPU in Real-World Test — nirurin · 2026-07-13