Do I need to address this issue at all? Thank you.
TrueNAS-25.04.2
There isn’t enough information to go on. What is your full hardware listing.
@HoneyBadger love the weekly show, any ideas about this please? Thank you.
Weird. Do you have any custom containers (eg: your Ollama install) or other Docker deploys that might have brought their own nvidia binaries to the party?
Yes, I’m running CodeProject.AI in a container for our CCTV at home which uses the GPU for A.I. tasks. I removed all the nvidia packages but now I can’t use the GPU for A.I. tasks, which I would ideally need. I still have one GPU assigned to the Container. How do I go about using the GPU in this container now? Thank you.
P.S. Where do you guys get packages for nvidia driver version 550.142 or are you installing it via .run file?
25.04.2 was still using Incus for the containerization system, IIRC, and things will be changing a fair bit in 25.10 back to a libvirt-lxc driven solution. I’m not sure if that will change things, but by “removed all the nvidia packages” you mean from the container, right? Or did you remove them from the TrueNAS base installation?
It might be that the CodeProject.AI container is build to target that specific runtime/version, and the earlier 550’s aren’t compatible because of bundled CUDA modules or similar. You might still have the same problem with 25.10 because we went to 570.172.08 and I think CUDA changed in 580 again.
We build the NVIDIA drivers at compile time into a systemd-sysext - some other community members have experimented with updating or changing the drivers - NVIDIA compatible driver test for TrueNAS 25.10 “Goldeye” - #16 by JeremC82
“removed all the nvidia packages” - from inside the container.
BTW, that’s really great what that member is doing with the nvidia mod, I’ve already started selling my P6000’s though and one has sold. With that money I bought a 2080Ti…timing aye! 
I can only guess that there’s some NVIDIA client piece that’s baked into the container - a kmod package, maybe? - that’s still hanging onto that 580.xx driver version.
There’s a container.ai docker image that I may try out, but currently I’m in the rather odd position of having no available NVIDIA GPUs to test this out with.