Edit: I used the laptop w/o external screen for a moment and no crash occurred. And I’m talking about the same workflows that caused 4 crashes in a row within 15 minutes with the monitor. @Mario_Limonciello I really need some input from someone who knows the internals, whether this is worth investigating. I may cut a case on amd freedesktop bug tracker if that’s desired way of moving forward.
This is nuts. Keeps happening, this time ChatGPT seems to trigger it super aggressively. My laptop is nothing but a very picey brick atm.
I even tested my memory with memtester to make sure it’s not damaged in any way.
I intend to open amdgpu bugtrack error with detailed description of my case. I’m triple-checked whether I have all the mitigations in place and I still suffer the problem. I don’t know what else to try at this point - different monitor? Not connecting monitor via Thunderbolt? It doesn’t seem like monitor transfer problem though.
If that doesn’t help I’m opening Framework support case.
If that doesn’t help I’ll strongly consider dropping this hardware. It’s disaster.
There’s a inrush of reports of stability issues on the system running new kernel, mesa and libdrm (if I interpreted it correctly all the distros involved in reports are Arch-based) on AMD bug tracking system. I have an impression that it almost always happen to people with quite good monitors - high res, high refresh rate. Personally I spent weekend in a place where I was plugged into older monitor without any fancy functionalities and it was OK.
KDE/Plasma was also reported in few cases. What gets to me is that kwin likely enables freesync by default. I just tried to disable it on both screens and gonna give it a go. Also as next experiment I’ll try to connect via HDMI, not DP (or its variations).
Edit 1: disabling VRR (either in settings or even in monitor) doesn’t help. Using only external screen also doesn’t help, it still triggers. HDMI next (and maybe rc kernel).
Given the huge number of reports coming in about this recently (I personally have a FW16 AI 300 with no GPU but with a high refresh rate high resolution monitor and KDE and am getting the same issue), have you tried downgrading your kernel to pre-6.18? I did so and haven’t had any issues for a little bit (though I didn’t have much luck reliably reproducing the crash before either so it might just not have happened yet).
Fair enough. Reason I bring it up is that at this point it’s quite clear that this isn’t a Framework issue (lots of the people reporting it are on other AI 300 laptops), so I don’t imagine Framework nor this forum can do anything about it; thought I’d mention a possible fix for others experiencing the issue who might see this thread or the other:
I see your point. Maybe this thread is indeed a bit excessive.
My only intention is knowledge sharing and building a database of potential solutions, obviously with hope to find viable one. That’s why I periodically update the first post to keep it tl;dr.
Speaking about that, I see that I forgot to mention that I have tried amdgpu.mes=0 with no luck.
I tried older kernels, back to 6.16. It didn’t help. My attention is more on mesa now. I tried downgrading it to 25.2 series, but then system becomes borked.
I’m installing 26.0.0 in hope for a fix. I’m quite hopeless at this point. Update: no, 26.0.0 doesn’t help, it still keeps hanging while I browse Google Photos (and watching YT video in background, not sure if it’s needed but I’ve found that this combination hangs it within 2 minutes).
Support won’t help until I use supported distro. I guess it’s fair - if the bug is real and 26.0.0 gets to Fedora/Ubuntu release, we may see more issues and pressure to fix it though. There’s always a possibility that I’m having some very, very peculiar setup. Even if so - the bug is still real and something triggers it, more or less often.
I just tried Fedora KDE livecd. All it takes is to install brave, login to Google and start browsing photos. It always happen at the moment when photos are sliding and blending together. Usually takes less than a minute to repro, longest I recorded was ~3 minutes.