Cannot get to work on AMD system with ROCm 6.2.3 on WSL (clean install) #6622
Description
Your question
I had ComfyUI working previously, but could not get some modules (e.g. F 5 TTS) to work no matter what I tried. I then deleted that install of Ubuntu and reinstalled a fresh copy of Ubuntu-22.04 in WSL using wsl --install -d Ubuntu-22.04
. I have not been able to get ComfyUI to work since.
I would like to get it working again because it is about 3x faster than in Windows 10.
I am a noob right now because I am just getting back into using linux, but I've put 30+ hours into figuring this out and not how to proceed. I've even attempted using LLMs to help troubleshoot to only some success.
If I follow the instructions/guide from this github, it doesn't work because it is looking for NVIDIA. This doesn't change when I install ROCm or point to the ROCm urls instead of cuda. I unfortunately didn't catalog my steps when I first got it working, but I was using python 3.12 and was not using venv. So far I've been trying to use Python 3.10, as it is specified in the AMD docs.
I've tried following the guide from ComfyUI-Manager github, but that also results in errors and cuda packages replacing the existing installed ROCm packages. This happens even if I use python -m pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/rocm6.2
Sys | Specs: |
---|---|
OS | Windows 10 Pro |
RAM | 96 GB |
VRAM | 36 GB |
CPU | AMD 7800X3D |
GPU1 | 7900XTX |
GPU2 | 6700X |
References used at minimum:
https://github.com/ltdrdata/ComfyUI-Manager
https://rocm.docs.amd.com/projects/radeon/en/latest/docs/install/wsl/install-radeon.html
https://rocm.docs.amd.com/projects/radeon/en/latest/docs/install/wsl/install-pytorch.html
https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/native-install/ubuntu.html
https://rocm.docs.amd.com/projects/install-on-linux/en/latest/install/prerequisites.html
https://rocm.docs.amd.com/projects/radeon/en/latest/docs/compatibility/wsl/wsl_compatibility.html
https://repo.radeon.com/amdgpu-install/6.2.3/ubuntu/jammy/
https://repo.radeon.com/rocm/manylinux/rocm-rel-6.2.3/
https://download.pytorch.org/whl/
https://download.pytorch.org/whl/rocm6.2
Steps:
wsl --shutdown
wsl --unregister Ubuntu-22.04
wsl -l -o
wsl --install -d Ubuntu-22.04
wsl -u [username]
The last command above enters the cli in Ubuntu without root as expected.
Then I follow steps to install ROCm at https://rocm.docs.amd.com/projects/radeon/en/latest/docs/install/wsl/install-radeon.html. Then do the same for PyTorch and Triton.
git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager comfyui-manager
I've tried running the follow before and after setting up a venv, and outside and within the venv, but it always breaks the existing rocm installs for the whole system—and I'm not forgetting source /venv/bin/activate
.
python -m pip install -r requirements.txt
python -m pip install -r custom_nodes/comfyui-manager/requirements.txt
When trying to run ComfyUI after this, it indicates several files cannot be found. I can still get valid output from rocminfo
, but the ROCm checks from https://rocm.docs.amd.com/projects/radeon/en/latest/docs/install/wsl/install-pytorch.html fail.
python -c 'import torch' 2> /dev/null && echo 'Success' || echo 'Failure'
python -c 'import torch; print(torch.cuda.is_available())'
python -c "import torch; print(f'device name [0]:', torch.cuda.get_device_name(0))"
I may be missing some details, so let me know and I'll provide.
Logs
Other
No response