-
-
Notifications
You must be signed in to change notification settings - Fork 454
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[Issue]: crash using Stablity Matrix on Fedora 41 with AMD GPU #3731
Comments
gfx1102 / RX 7600 series are unsupported by ROCm, you have to downgrade to ROCm 5.7. Also why are you using |
Zulda is turned on because Stability Matrix turns it on by default. Switching to ROCm has identical results with not installing with the exact same error message. This exact hardware has worked before on this operating system in December - and works on other Linux distros, even with ROCm 6.2 - and doesn't anymore. Something changed with an update to either Fedora or SD.Next around the New Year, and it broke. if it's Fedora I can figure it out on my end, but I need to know what the problem is in order to fix it and Error 139, from what I recall, is "segfault" which is a catchall for "Whelp, it just crashed" and from what I can see the log doesn't actually say anything useful. The thing is, I've read the exact opposite and that you MUST have a 6.x branch of ROCm to get the hardware to work right. I honestly don't think anyone actually know what they're talking about, everyone has contradictory messages, and I include the engineers at AMD developing ROCm. |
first thing that happens after last line in the log is setting some torch parameters. |
@panton41 any updates? |
I am working on trying things right now. The last week has been busy at work and I haven't had a chance to do anything AI related at home. |
So, there's two things going on. One, the install process in Stability Matrix crashes with the previous error. But, on reopening Stability Matrix it shows as installed and returns an error that it was throwing back in early January when the whole problem with it not working began. The relevant part of the kernel dump:
The directory Stable-Diffusion/Data/Assets/ is related to the custom Python libraries Stability Matrix installs. The new log:
The other error from when it seems to be installed and fails to run is:
Which began at the same time the previous error started to happen so I imagine they're related in some way. |
so its crashing inside rocm. this can be due to any number of reasons, but its is typical for amd if its attempting to use wrong functions intended for wrong gpu subtype. search previous issues for |
It detected the gpu fine, the issue is this:
|
But, repeating yet again, this was working back in December, on a Fedora 41 derivative, and stopped working in early January with these exact errors. Fedora 41 shipped with ROCm 6.2, so it was working until that point. Edit: From the looks of it while searching, AMD considers the incompatibility a regression and seems to be working on fixing it. Breaking compatibility on their latest generation of parts is a bad look, but AMD seems to be masters of snatching defeat from the jaws of victory when it comes to competing with nVidia so it seems par for the course. |
Issue Description
I had SD.Next running on Bazzite Linux (which is based on Fedora 41) until about three weeks ago when this current error began. I changed distros to Linux Mint and had no problem with SD.Next, but there were issues in other programs, so I bounced around until I was on Fedora 41 itself and the previous problem continues. I am unsure whether his has to do with a Fedora update or an update to SD.Next, but I've been unable to get previous versions installed as well with the same error.
During the installation, an exception is thrown, and the process stops.
Stability Matrix is an App Image that creates an isolated environment to install various Stable Diffusion programs without worrying too much about system libraries.
Version Platform Description
Fedora Linux 41 x86_64
AMD Ryzen 5600X
64GB DDR4
AMD Radeon 7600 GT
No logs are created that contain the application version because it fails to install.
Relevant log output
Backend
Diffusers
UI
Standard
Branch
Master
Model
Other
Acknowledgements
The text was updated successfully, but these errors were encountered: