[Help] Getting a noise instead of an image

[milutinke]

Hello,

I am completely new to image AI generation and stable diffusion. I've installed the ComfyUI and the ComfyUI-Manager and downloaded these 2 models, one during the installation, the other one with the ComfyUI-Manager Edit: Installed more models, this happens on any model.

I run arch linux (zen kernel) with Red Devil RX 6800 XT OC 16 GB, 16 GB RAM and i5 11600k, I have the latest mesa-git driver for AMD installed.

When I try to generate an image, I get either a noise like this or a black image Can someone tell me if I am doing anything wrong or is this a bug or a problem with my configuration/hardware ?

[comfyanonymous]

Try installing the latest ROCm pytorch: https://github.com/comfyanonymous/ComfyUI#amd-gpus-linux-only

[milutinke]

Try installing the latest ROCm pytorch: https://github.com/comfyanonymous/ComfyUI#amd-gpus-linux-only

I've tried with the one recommended, and also with the nightly build, it's the same

[comfyanonymous]

Try running it with: --force-fp32

[milutinke]

Try running it with: --force-fp32

It's still the same. I've tried it also with the newer ROCm but it crashes then

[ltdrdata]

Have you been checked through --cpu mode?

[milutinke]

Have you been checked through --cpu mode?

It works, but it's very slow, and I'd like to use a GPU if possible. Any other ideas?

[Patryk27]

In my case (VEGA 64 on passthrough) this turned out to be caused by missing support for PCIe atomics in QEMU - installing this patch helped and now both stable-diffusion and ComfyUI work perfectly.