[Bug]: "Torch is not able to use GPU" on RX 580 2048SP
Is there an existing issue for this?
- [X] I have searched the existing issues and checked the recent builds/commits
What happened?
Trying to run unmodified WebUI on Ubuntu 22.04 (i5-10500 + RX570 8GB), does not run without --skip-torch-cuda test, can use only CPU.
Steps to reproduce the problem
- Clone WebUI
- Put any model in models/Stable-diffusion
- Launch WebUI with TORCH_COMMAND='pip install torch torchvision --extra-index-url https://download.pytorch.org/whl/rocm5.1.1' python launch.py --no-half
What should have happened?
Successful launch of WebUI; RX500 series should have no problems.
Commit where the problem happens
b08500cec8a791ef20082628b49b17df833f5dda
What platforms do you use to access the UI ?
Linux
What browsers do you use to access the UI ?
Mozilla Firefox
Command Line Arguments
No
List of extensions
No extensions, unmodified webui
Console logs
https://pastebin.com/qpzsCS6K
Additional information
Rocminfo successfully detects the GPU as RX 580 2048SP, but I am unable to use it.
I thought you can only use NVIDIA gpu and not AMD?
AMD can be used. And documentation says that RX500 should have no issues and can run WebUI without --no-half and --precision full
I have confirmed that rocm and opencl actually works on this gpu.
rocminfo:
clinfo:
UPD: I assume custom torch is needed now, but i couldnt find one that will install on ubuntu 22.04 (wheel is not supported on this platform) https://pastebin.com/e9dPQwWV
UPD: I assume custom torch is needed now, but i couldnt find one that will install on ubuntu 22.04 (wheel is not supported on this platform) https://pastebin.com/e9dPQwWV
I used these ones, works with my rx584 and kubuntu 22.04 both with auto and voldy fork: https://github.com/tsl0922/pytorch-gfx803/releases
UPD: I assume custom torch is needed now, but i couldnt find one that will install on ubuntu 22.04 (wheel is not supported on this platform) https://pastebin.com/e9dPQwWV
I used these ones, works with my rx584 and kubuntu 22.04 both with auto and voldy fork: https://github.com/tsl0922/pytorch-gfx803/releases How did you install these? Installed them in venv?
How did you install these? Installed them in venv?
Yes, activated venv and installed from there I also manually installed some missing libraries giving errors, libstdc++-12-dev, openmpi and libopenblas, if i remember correctly
okay, will try, not closing yet
okay, will try, not closing yet
@marohinmark So what did you do to make it work? I'm getting a similar error on my RX 580 and have rocm 5.3 installed
Have same issue =(