I use Arch Linux as OS and I have my AI stack (ComfyUI, llama.cpp, n8n, Open WebUI, Perplexica) running in Docker containers.
Desktop
- don’t update/upgrade to Bios 3.04 because it is some kind of broken
- Framework Desktop Ryzen AI MAX 300 BIOS 3.04 Release STABLE
- FW Desktop 3.04 slow grub/linux boot · Issue #138 · FrameworkComputer/SoftwareFirmwareIssueTracker · GitHub
- it runs stable
- it still has some kind of USB issues as you can read in other threads and I also experience them some times at boot due to my system build
ComfyUI
- is based on an offical AMD ROCm Pytorch image with some additional parts (rocm/pytorch:rocm7.1.1_ubuntu24.04_py3.12_pytorch_release_2.9.1)
- Z Image Turbo runs great
- WAN 2.2 5B works but is slow, likely due to way to less compute power of the APU compared to big dGPUs
- WAN 2.2 14B results in OOM, very likely a software issue on Pytorch/ROCm/… side
llama.cpp
- is the official vulkan image (Package llama.cpp · GitHub)
- router mode is currently some kind of broken for any longer running jobs
n8n
- is the official image (docker.n8n.io/n8nio/n8n)
Open WebUI
- is the official image (ghcr.io/open-webui/open-webui:main)
Perplexica
- is the official image (itzcrazykns1337/perplexica:latest)