Can't pass specific GPU to app

[derp90]

I have Nvidia GPUs in my system
A Quadro p600 (in a x1 slot) That is used as a display out only. Ie for the native console out, bios, and GUI
A Tesla P4 that should be used for apps (does not have display plugs)
When I do "Allocate 1 nvidia.com/gpu GPU" only the p600 is visible in apps (Specifically plex right now)
When I do "Allocate 2 nvidia.com/gpu GPU" I get an error in the kube log

Code:

Allocate failed due to rpc error: code = Unknown desc = request for 'nvidia.com/gpu: 2' too large: maximum request size for shared resources is 1, which is unexpected

Is there a way to fix this without isolating a GPU? Isolating the GPU disables the ability to see the built in console of truenas.

Also, why is the GPU menu for apps not the same menu as the isolating GPU menu so that we can choose what gpu(s) to pass?

 

[derp90]

Slight update, restarting the container seems to just pull a random gpu, as the P4 is now being passed to plex.

 

[farout]

Are you isolating the GPU that you want to use with the Plex app ?
If yes, then this is the problem, since an Isolated GPU will not be available for Truenas and its apps anymore.
Isolation is only for VMs.

 

[derp90]

Are you isolating the GPU that you want to use with the Plex app ?
If yes, then this is the problem, since an Isolated GPU will not be available for Truenas and its apps anymore.
Isolation is only for VMs.

No GPUs are isolated

 

[HoneyBadger]

Hey @derp90

Can you file a feature request for this through Report a Bug at the top of the forums? We'd probably need to make adjustments to the middleware to integrate with libnvidia-container to get GPU UUIDs (and make them more human-readable) for selection on a per-app basis.

 

[derp90]

Hey @derp90

Can you file a feature request for this through Report a Bug at the top of the forums? We'd probably need to make adjustments to the middleware to integrate with libnvidia-container to get GPU UUIDs (and make them more human-readable) for selection on a per-app basis.

Done!

 

[HoneyBadger]

Thanks, got it as NAS-126873 - if you can attach a debug to it when the Jira system prompts you that will probably help out on our end. :)

 

[derp90]

Thanks, got it as NAS-126873 - if you can attach a debug to it when the Jira system prompts you that will probably help out on our end. :)

What would I attach a debug log from?

 

[HoneyBadger]

What would I attach a debug log from?

From your TrueNAS system, the one with the multiple GPUs that isn't working as desired. Use System -> Advanced -> Save Debug and then attach it via the "Private/Debug Upload" link that was sent to you.

 

[derp90]

Done!