I debugged this issue using nvidia forums + fedora + KDE and lastly with Claude
Claude had all the info, I made it write this write up since me writing it would be worse and I would have forgetten something.
I checked that all the info here is accurate. Please if you have a solution I’d appreciate it.
System Info:
- GPU: NVIDIA GeForce RTX 4070 Ti
- Driver: 595.71.05
- OS: Fedora 44
- Kernel: 7.0.9-202.fc44.x86_64
- KDE Plasma: 6.6.5
- Display Server: Wayland (KWin)
Monitor Setup:
- DP-1: AOC Q27G4ZR (2560x1440 @ 240Hz)
- DP-2: LG UltraGear (2560x1440 @ 164Hz)
- HDMI-A-1: LG UltraGear (1920x1080 @ 144Hz)
Description:
On cold boot with all three monitors connected, KWin Wayland fails with DRM atomic modeset errors and the session crashes back to the login screen. Once in this failed state, warm reboots also fail.
The bad GPU state persists across reboots until a monitor is physically unplugged and replugged, or the GPU is fully power-cycled by unplugging a monitor while the PC is off.
The only reliable recovery without hardware intervention is catching the brief KDE session flash during the failed boot and immediately clicking reboot before being kicked back to the login screen.
Key log errors (journalctl -b -1):
kwin_wayland: Atomic modeset test failed! Permission denied
kwin_wayland: Applying output configuration failed!
nvidia-smi data across all three states:
| Column 1 | Column 2 | Column 3 | Column 4 |
|---|---|---|---|
| State | P-State | Power Draw | Power Cap |
| Working (2 monitors) | P5 | 17.02W | Not Active |
| Failing (3 monitors, stock) | P8 | 13.49W | Not Active |
| Failing (3 monitors, PerfLevelSrc=0x2222) | P0 | 30.82W | Not Active |
The P0 result with PerfLevelSrc=0x2222 confirms the failure is not power-state related, the GPU is at maximum performance but still cannot complete the atomic commit for 3 outputs. This points to a pure DRM atomic modeset sequencing bug in the driver when handling 3 simultaneous outputs.
Reproduction steps:
- Connect 3 monitors as listed above
- Boot normally and log in — session works correctly
- Fully shut down
- Cold boot and attempt login
- Session crashes back to login screen after ~30 seconds
- Subsequent warm reboots also fail until a monitor is unplugged
Confirmed workarounds:
- Unplugging any one monitor before cold boot, then plugging back in after login
- Plugging the third monitor in while the PC is fully off before first boot (sometimes works)
Confirmed NOT the cause:
- SELinux (tested with setenforce 0)
- Power state (fails even at P0 with PerfLevelSrc=0x2222)
- kwinoutputconfig.json corruption (files byte-identical between working and failing boots)
- Refresh rate mismatch (fails with all three monitors at 144Hz)
- KDE Plasma version (tested on 6.6.5 which includes NVIDIA 595 fixes)
- nouveau driver conflict (blacklisted, confirmed not loaded)
- Driver reinstall (clean reinstall performed, issue persists)