Hi. SCALE 25.04 installed. Ripped out working Nvidia P2000 today, put an Intel A380 in instead. I have Dockge installed to manage containers.
Went into the Dockge interface after the hardware swap, and was met with Error response from daemon: unknown or invalid runtime name: nvidia when attempting to launch _traefik_ (which of course has no mention of nvidia).
I inspected /etc/docker/daemon.json, again no reference to nvidia. I then did some cross-checks using ChatGPT and it suggests Docker is hooked into /usr/bin/nvidia-container-runtime and thus this must be disabled. Via a “shadow” file since /usr/bin is mounted R-O!
This sounds.. ridiculous. Could some kind soul offer a more sensible way forward? None of my containers now launch, and very few even mention nvidia!
That appears to have worked - thank you! I’m guessing this is a Docker bug?
Curiously - and in case anyone else experiences this - one container still persisted with needing nvidia - Postgres! Which needless to say never mentioned nvidia in the first place - it seems it did launch in the background however Dockge didn’t spot it and came up with this older error instead. Managed to find Stop and Inactive under the Update drop-down, used it (resulting in a stopped stack) then redeployed and it worked!
Allow me to follow-up on the underlying task I had - the switching from nvidia to Intel Arc hardware for Plex transcoding purposes (at least).
The solution from @LarsR works to get containers running again - thanks again. I’m guessing there’s a persistence somewhere in runc perhaps that shouldn’t exist. Moving on though…
For the Plex container, while Plex showed the Intel Arc in the drop-down as a hardware transcoding device, it absolutely refused to use it. How do I know? Check the dashboard inside Plex while streaming something and changing the stream to a downscaled version (go from a 4k file down to 1080p by asking within the Plex client). For me, the CPU use rocketed. Also, inside the Plex logs I although during boot it spotted my Intel Arc device, it then errored when the transcoding began:
No VA display found for device /dev/dri/renderD128
Google for this and you’ll get talk about permissions. You can’t follow the advice to add the Plex user within TrueNAS to the render group though - not permitted. Instead, I spotted a fragment from an AI result to the above error which didn’t come from the referenced source:
Removing the /dev/dri from volumes where it was working file with Plex and instead passing it through as a device solved the problem. Well done to Brave’s AI for that.
And, FWIW, this appears to work without the other suggested solutions found online of:
Pass group_add of 107 (render) through via docker-compose. I tried it, made no difference, so I removed it.
Switching off HDR Tone Mapping. I left this enabled, no problems yet.
