[SOLVED]Cannot assign NVIDIA GPU to App(Plex) after Cobia

fayelund

Dabbler
Joined
Aug 27, 2021
Messages
31
After upgrading to Cobia I cannot pass my GPU to Plex app anymore.
This worked fine in Bluefin(last version: 22.12.4.2)
On the drop sown menu there is no GPUs available to choose.

nvidia-smi sees the GPU:
1698309039588.png
 

fayelund

Dabbler
Joined
Aug 27, 2021
Messages
31
In Bluefin isolation was only used when assigning GPU to VMs, not apps.
But I have of course tried that as well, on Cobia, with no luck.
 

probain

Patron
Joined
Feb 25, 2023
Messages
211
"You can no longer share a GPU between apps since docker has been removed. A little disappointing, but that's really an edge case..."
Source - Reddit

Maybe that is related?
 

fayelund

Dabbler
Joined
Aug 27, 2021
Messages
31
"You can no longer share a GPU between apps since docker has been removed. A little disappointing, but that's really an edge case..."
Source - Reddit

Maybe that is related?
Since there are drop down menus to assign GPUs from Nvidia, AMD and Intel in the new apps interface I doubt that. Why create them in the new interface if the option was removed?
 

probain

Patron
Joined
Feb 25, 2023
Messages
211
Since there are drop down menus to assign GPUs from Nvidia, AMD and Intel in the new apps interface I doubt that. Why create them in the new interface if the option was removed?
I agree. Have you filed a Jira ticket for it? It might just be a simple bug. Or something else entirely.
 

fayelund

Dabbler
Joined
Aug 27, 2021
Messages
31
I agree. Have you filed a Jira ticket for it? It might just be a simple bug. Or something else entirely.
I have not yet. Figured I would probe the community first to see if there was a simple solution, but the time to file a ticket is getting close :)
 

fayelund

Dabbler
Joined
Aug 27, 2021
Messages
31

fayelund

Dabbler
Joined
Aug 27, 2021
Messages
31
Did you get a solution to this, Im having the same issue. Look like the ticket does not exist?
The ticket got marked with "fix version 23.10.1", so if I remember correctly mid December
 
Last edited:

fayelund

Dabbler
Joined
Aug 27, 2021
Messages
31
Appears this had been resolved with app updates.
No not resolved at all on my side. Just checked, no updates to app and not possible to assign GPU.
 

Mortorojo

Cadet
Joined
May 25, 2023
Messages
8
This is odd, It was just working for me. Saw someone elses post on the same thing, checked app setting and it was availible, updated the app and it was working. Restarted truenas and now its no longer availiable.
 

joace

Dabbler
Joined
Mar 15, 2014
Messages
11
The same on my system, the latest Cobia 23.10.0.1 and all apps updated on 11/10/23, if ticket said fixed in 23.10.1, probably have to wait for some tiems.
 

Mortorojo

Cadet
Joined
May 25, 2023
Messages
8
I found a workaround that might work for you. If you haven't updated your app, try updating and see if they show up. If you are upto date roll back one app, i.e. plex, i rolled back to before 11-5-23. At least for me that made the gpu allocateable, assign a gpu. Once again at least for me that enable the ability to assign gpu's on all apps. Then update the app that was rolled back and assiged a gpu. DO NOT REBOOT TRUENAS, this will break it again till they can fix the underlying issue.
 

joace

Dabbler
Joined
Mar 15, 2014
Messages
11
the issue on my system seems a little different, I can assign 1 gpu to both Plex and frigate, but no process seen using gpu in nvidia-smi, not the same issue of assignment reset, tried reboot system,assignment still there, but no process
 

Alimnette

Cadet
Joined
Jun 27, 2014
Messages
2
I have a same issue.
i can see gpu with nvidia-smi (P2000) and assign in container plex but anything process isn't use.
 

fayelund

Dabbler
Joined
Aug 27, 2021
Messages
31
This is now fixed on my side. My proxy seemed to block som updates for the Plex application.
 

Alimnette

Cadet
Joined
Jun 27, 2014
Messages
2
it's solved for me.

juste check if you have "runtimeClassName: nvidia" en manifest.
you can try this in shell: k3s kubectl -n ix-plex get deploy -o yaml | grep runtimeClassName
if you dont have "runtimeClassName: nvidia" then add it by:
"k3s kubectl -n ix-plex edit deploy"

plex.png


after => escape => : => wq => enter

wait and check :)
 
Top