System
- GPU: RTX 4070 Ti SUPER (PCI 0000:05:00), 16 GB
- Driver: 610.57.04
- iGPU: AMD (PCI 0000:11:00), unused for this session
- OS: Fedora 44, kernel 7.1.8-200.fc44.x86_64
- Desktop: KDE Plasma 6.7.4, KWin 6.7.4, Wayland
- Displays: 2x 2560x1440 @ 165 Hz, both on DisplayPort off the discrete GPU
- DP-4: LG ULTRAGEAR
- DP-5: Sentronic SKP_E30-27
- Night Light: disabled
Summary
After 7.5 hours of uptime, a single Xid 56 was raised while a hardware-decoded video was playing in a browser tab. From that moment the display engine never recovered: KWin emitted a pageflip timeout every second for the next 10 minutes until I power-cycled the machine. The rest of the GPU kept running and reporting normally throughout.
Timeline
08:57:00 boot
09:02:26 KWin session starts
16:06:24 last notable compositor activity (gwenview entering fullscreen)
16:32:32 NVRM: Xid (PCI:0000:05:00): 56, CMDre 00000007 00000000 00000000 00000001 00000000
16:32:33 kwin_wayland: Pageflip timed out! — repeats 1/sec, 621 occurrences
16:36:41 Krunner1Adaptor::Teardown — my Ctrl+Alt+T attempt; input and D-Bus still alive
16:42:56 [drm:nv_drm_atomic_commit [nvidia_drm]] *ERROR* [GPU ID 0x00000500] Flip event timeout on head 0
16:54:26 journal ends (hard power-off)
Exactly one Xid 56. No cascade, no repeat, no subsequent fault. Everything after it is KWin retrying a flip that never completes.
Context at the time of the fault
No game, no brightness or colour temperature change, no display hotplug.
A video was playing in a browser tab with hardware decode active. Decode continued normally while scanout was dead.
Browser stack:
- Vivaldi 8.1.4087.68 (Flatpak, com.vivaldi.Vivaldi), Chromium-based
- org.freedesktop.Platform.GL.nvidia-610-57-04 (exact match to host driver)
- org.freedesktop.Platform.VAAPI.nvidia 25.08
vivaldi://gpureports Video Decode: Hardware accelerated, Compositing: Hardware accelerated, Vulkan: Enabled
The VA-API stack was correctly configured and functioning — this is not a fallback or misconfiguration case.
System state during the hang
A GPU logger was running, giving 500 ms samples covering 16:41:31 to 16:42:31 — nine minutes into the hang, while flip timeouts were still being emitted:
| Metric | During hang | After reboot, under load |
|---|---|---|
| GPU / hotspot temp | 42 / 53 °C | 54 / 64 °C |
| Power draw | 11.5 W (285 W cap) | 71.6 W |
| Core clock | 210 MHz (P8) | 2700 MHz |
| Memory clock | 405 MHz, pinned | 10501 MHz |
| VRAM used | 1.6 / 16 GB | 1.5 GB |
The memory clock stayed pinned at the 405 MHz idle state for the entire duration, while two 1440p165 outputs were still nominally active — bandwidth that cannot sustain scanout for those modes.
The session itself was never lost: the Krunner D-Bus teardown at 16:36:41 shows input events and D-Bus were still being processed four minutes after the fault. Only the display was gone.
Thermal, power delivery, VRAM exhaustion and memory pressure are all excluded by the telemetry above — the GPU was in deep idle at 42 °C drawing 11 W of a 285 W cap. glmark2 after reboot scores 16747 with no artefacts, so the card is functionally healthy.
Recovery attempts, all unsuccessful
- Unplug/replug DisplayPort → black screen
- Moving a display to the motherboard DP output → nothing; the amdgpu side did attempt DP link training at 16:45:57 but the compositor was already wedged
- Ctrl+Alt+T → no window ever drawn
sudo reboot→ the journal ends with no shutdown sequence, so it never executed
Hard power-off was the only way out, 22 minutes in.
Attached
nvidia-bug-report.log.gz— generated on the following boot; verified to contain the Xid and flip timeout from the previous bootkwin-prevboot.txt—journalctl --user-unit plasma-kwin_wayland -b -1
nvidia-bug-report.log.gz (551.2 KB)
kwin-prevboot.txt (224.4 KB)