Remix.run Logo
▲ throwaway2037 a day ago

    > A few months ago I threw Claude at the problem and it took a few diagnosing sessions, but the workarounds it eventually applied have been holding, luckily, so all's been well for a while.
What was the fix?
▲skeledrew a day ago | parent [-]

It made a service that monitors journalctl for messages that indicate amdgpu is erroring, and does a few things like notify me that I should save in case the error isn't recoverable, restart the window manager, reset the GPU via debugfs and more. Also set amdgpu.dcdebugmask=0x410 in grub. And wrote a bug report for Ubuntu, Framework and/or amdgpu driver repo, but I never submitted as generally too many complaints going around about slop issues; it's fine long as things work for me.

▲throwaway2037 14 hours ago | parent [-]

Cool story; thank you to share the details. It is pretty wild to me how "smart" LLMs are these days to debug technical issues. Honestly, it scares me a little bit!

Is your plan to never upgrade the BIOS again to keep your system as stable as possible... or will you take the gamble on future BIOS upgrades hoping it will fix your issue?

▲skeledrew 13 hours ago | parent [-]

Yep they're good at many things. I have over 2 dozen sys admin write-ups (I have Claude do a write-up each time it does some troubleshooting, etc so I can always go back and check and it always has a reference for anything reoccurring) so far, and my system has been a lot better off.

I still do upgrades as if something breaks again I can easily have it handled, as long as there's internet connectivity. And it's possible the issue is already handled upstream; no way to know as it always occurred randomly, though more often while I was working on a project which I guess might have been pushing the GPU.