Plex Media Server was repeatedly crashing immediately after startup with SIGSEGV / core dump. The Plex database checked clean, and Plex would start successfully only after hiding CUDA from the Plex service.
Root cause appears to have been an NVIDIA driver/library version mismatch. The userspace CUDA/NVML libraries and the loaded NVIDIA kernel module reported different versions. I do not know what caused the mismatch.
A reboot cleared the mismatch, after which the NVIDIA stack reported consistent versions and Plex started normally again.
Environment
- Plex Media Server version:
1.43.1.10611-1e34174b1 - OS: Ubuntu 24.04 / Noble
- NVIDIA driver branch involved:
580 - Plex install path:
/usr/lib/plexmediaserver - Plex application support directory:
/var/media/media-6/plexAppDir/Library/Application Support/Plex Media Server
Symptoms
systemctl status plexmediaserver showed Plex repeatedly failing with a core dump:
plexmediaserver.service: Main process exited, code=dumped, status=11/SEGV
plexmediaserver.service: Failed with result 'core-dump'
plexmediaserver.service: Start request repeated too quickly
Failed to start plexmediaserver.service - Plex Media Server
If it keeps happening I can ignore it with adding
[Service]
Environment="CUDA_VISIBLE_DEVICES=-1"
to
sudo systemctl edit plexmediaserver
What I would expect Plex to have done would be to disable/fallback to not using gpu. Bonus if it were to notify me in the dashboard that there is a problem.
