Plex Nvidia M2000 Container Issues

Hello,

I have been using a VM with GPU PCIe passthough and docker for quite some time and my intention has been to migrate plex to SCALE.

Yesterday I turned up plex locally using my docker compose yaml (not the scale app) and started to experiance issues when using hardware transcoding. I just get a black screen and a paused stream and under nvidia-smi, card is visable, but does not show any executions.

After troubleshooting today, I’ve found the 550 drivers used by 24.10.1 are not happy with my Nvidia M2000. I installed 550 (550.120, 550.127.05 where not available for my VM distro) on my VM and experianced the same issues.

Is there a way to force a specific version of the nvidia drivers?

Thanks,

In case it’s related, the latest update to EE 24.10.1 seems to have broken my GPU setup. I’ve only just discovered it and come here to find out where /dev/dri is now, as it seems to be missing…

I am also on 24.10.1, but do have /dev/dri/card0… If that helps in anyway :person_shrugging:

edit:
ok, so I tested this further and that card was not my M2000, based on the by-path and “midclt call app.gpu_choices | jq”.

@marshalleq I think this is linked and additionally possibly liked to nvidia_drm.

I would like to do some additional troubleshooting when I have time. As such, I’d like to disable rootfs protection, does anyone know how to re-enable this when no changes have been made?

/usr/local/libexec/disable-rootfs-protection

There is no enable scripts.

Thanks. I guess its just dropped the GPU for some reason.

So update 24.10.2 fixes the nvidia_drm module loading… yay… however…

Plex runs and the Nvidia card is visable, I even see the process execute using nvidia-smi, but I get a black screen still, but if I wait a while, it starts hw transcoding. I see a spam of errors " Invalid data found when processing input".

I do not get this delay with my VM or the error spamming.

Any ideas?