AMD's Ryzen AI Max+ 395 (Strix Halo) has been on sale since early 2025, giving AMD over a year's head start on Nvidia's newly announced RTX Spark laptop platform. At Computex, AMD executives pointedly welcomed Nvidia to a market they've been building. On specs, both platforms top out at 128GB unified memory, though Nvidia claims slightly higher memory bandwidth and FP4 compute advantages. The real story is ROCm's maturation: PyTorch now treats AMD as a first-class accelerator target, llama.cpp, Ollama, LM Studio, and ComfyUI all run on AMD hardware, and installation has become far less painful. Gaps remain — CUDA still leads in ecosystem breadth, day-zero software support, training workloads, and low-precision matrix math — but ROCm is no longer the reason to avoid AMD for local LLM work. AMD's Ryzen AI Halo developer mini-PC ($3,999) also directly targets Nvidia's DGX Spark with a pre-validated software stack.

12m read timeFrom xda-developers.com
Post cover image
Table of contents
AMD has been selling this class of machine since early 2025The spec sheets are closeROCm has quietly become usable for the software people actually runROCm still trails CUDA in specific ways
71 Impressions