Transcoding intermittent issue

[thunderb0lt]

That’s a clever workaround. It’s not an ideal solution, but at least it keeps things running. Have you looked into whether the GPU is being reset or if there’s a driver issue? Maybe checking dmesg or nvidia-smi logs could provide more clues. Also, does the issue happen under specific workloads, or is it random? Curious if there’s a more permanent fix.

[_Nick]

Hi,

I'm bumping this thread as I'm experiencing the same intermittent issue, and no reliable fix or root cause has been identified. It’s also under-discussed and may be fuelling the perception that Nvidia GPUs are unreliable for Docker-based Jellyfin transcoding.

Issue:
Transcoding works flawlessly for days or weeks, then suddenly stops working with users experiencing a fatal error immediately after triggering any transcoding:
Dashboard logs show: 

Code:

CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected

Rebooting the system always restores functionality. Restarting just the Jellyfin Docker container (on Ubuntu) does not resolve it, only a full system reboot does, after which transcoding works again for extended periods.

Testing:
I'm currently testing whether unattended upgrades affecting the Linux kernel are the cause. I've set a reminder to update this post in 4–5 weeks if disabling unattended upgrades works.

Regards,
Nick