AMD Ryzen AI Max+ Runs ComfyUI Natively on Windows

AMD published a new setup guide this week showing how to run ComfyUI directly on Windows using a Ryzen AI Max+ processor, skipping the Windows Subsystem for Linux step that's long been the default path for AMD-based image generation work. The change hinges on ROCm 7.2.1, which now supports a native PyTorch build on Windows, letting the software talk straight to the integrated Radeon 8060S GPU without a virtualization layer in between.

That matters for anyone running SDXL, Flux.1-dev, or video models like WAN 2.2 locally, since those workflows tend to be memory-hungry. The Ryzen AI Max+ chip's unified memory architecture lets its integrated GPU pull from a much larger pool than a typical discrete card would offer on its own, which is the real selling point here for creators trying to run bigger models without a separate GPU.

AMD's post follows earlier guides the company put out for discrete Radeon RX 9000 series cards on Linux, but this one is aimed squarely at people working natively on Windows with integrated graphics. The walkthrough covers installing ROCm and PyTorch, binding ComfyUI to the 8060S, and running actual generation jobs to see how the setup performs.