Hardware:
- Framework Laptop 13, AMD Ryzen AI (“Strix”), integrated AMD Radeon 890M (
amdgpu, PCI ID 1002:150e)
Software:
- Fedora 44 (Workstation/GNOME)
- Kernel: 7.1.3-201.fc44.x86_64 (installed same morning via normal updates)
- Mesa: 26.1.4 (radeonsi, ACO)
What happened:
After several hours of uptime (system had been running since that morning, ~9 hours in), my screen suddenly filled with heavy visual corruption: random colored pixels/artifacts all over the display, looked like a classic “GPU died” scenario (at least to what I would associate it with?):
At the time I had a fairly typical but non-trivial load: several Neovim instances, a couple of Copilot CLI sessions, and quite a few Chrome tabs open, plus the usual Electron apps (Discord, Slack, 1Password). I hard-rebooted and everything came back to normal immediately, no repeat issues since. Interestingly it was working when I had these artifacts, they would glitter when moving the mouse around, I even managed to power off the system using the OSs UI.
What I checked in the logs:
I went through journalctl for the affected boot in detail, specifically looking for:
- Kernel panics/oops
amdgpuring timeouts, GPU resets, VM faults- MCE / hardware errors, RAS/ECC errors
- OOM kills
Found nothing. The boot in question ended in a completely clean, user-initiated shutdown sequence, no amdgpu warnings/errors at all in the last 30 minutes before reboot, no ring hangs, no GPU reset events. This suggests the corruption was contained to the display pipe / compositor output (a DC atomic-commit or scanout-level glitch) rather than a full GPU engine hang, since a hard hang would normally trigger amdgpu’s reset/recovery path and get logged.
Other context that might be relevant:
- A suspend/resume cycle happened about 2.5 hours before the incident, which also logged two USB resume failures (
error -5) on that occasion. Not sure if related, but AMD iGPU resume-related display glitches are a known bug class on Strix/Strix Halo. - This chassis runs fairly hot under load: CPU (Tctl) was reading ~79°C shortly after the incident/reboot, and fan speed sits around 7000 RPM even under just moderate load (browser + editors + CLI tools). No explicit thermal-throttle/PROCHOT log lines were found, but AMD’s SMU can throttle silently without logging, so I can’t rule out a heat-related trigger.
- No kernel, Mesa, or amdgpu-firmware packages were part of any recent update — the only update near this event (which triggered an unrelated automatic reboot+poweroff, hence needing a manual power-button press) was a routine offline package update for unrelated apps (Firefox, VS Code, binutils, NetworkManager-ssh, etc.).
Question for the community: has anyone else seen sporadic display corruption like this on Strix/890M with amdgpu, especially under heavy multi-window/compositor load or after suspend/resume? Any known kernel/mesa combos that fix it, or should I just chalk this up to a one-off and keep an eye on dmesg/sensors next time?
Happy to attach specific kernel log excerpts or sensors output if useful — I have them saved.