Hi,
In Scale EE 24.10 there is the option to pass through Intel GPU to the apps when deploying.
In case where there are multiple Intel GPU’s available, like iGPU and dedicated ARC card, is there any way to specify which of these GPU will be used in that app ?
Thank you !
No, the way non-nvidia cards pass through there isn’t a clean way for us to split them up. Its basically the /dev/dri directory being shared, which exposes all cards.
Thank you for the answer.
Then is there a way to change their order ? The reason why i’m asking is that some apps would choose the first card available and are not able to use the other after the app has been deployed.
Particularly Frigate app.
Not that I’m aware of. Dockers GPU support for selecting is limited to nvidia only.
The way you share a non-nvidia GPU is by passing through the /dev/dri directory “wholesale” and we don’t have a way to let you pick and choose between devices from the host side. Inside the container if the App is sophisticated enough it might support letting you select from the available GPUs on that side.
ok. thank you. Hope the app devs will provide this option in the future.
Cheers !
Hi, I was just wondering how you implemented device passthrough in the docker image with the wholesale method.
I installed a random container (tdarr) that had the option of GPU passthrough. I just installed it to test this, so the option I changed was ticking the GPU passthrough box.
I ran sudo docker inspect --format='{{json .Config}}' ix-tdarr-tdarr-1 in the shell and there was no devices /dev/dri stuff seen. I am assuming I am missing something.
Also couldn’t you just do something like the following in docker compose?
devices:
- /dev/dri/renderD129:/dev/dri
Thanks
or potentially:
devices:
- "/dev/dri/card0:/dev/dri/card0"
- "/dev/dri/renderD128:/dev/dri/renderD128"
sorry not an expert on this, just seen other people getting it working like this.
Looks like the apps choose the first Intel GPU available.
I managed to trick it by:
This way the app actually chooses my second GPU (ARC card), not the iGPU.
Sorry deleted last post because was a bit a dumb.
Thanks for the quick reply. Sorry I should have been more specific with my question, I was asking kris about truenas implementation on this.
That is super interesting. Thank you mentioning that worked for you, will do testing with that sort of setup.
I have a server with 3 various ARC GPUs and a very basic GPU for Truenas, so this would be very useful for me.
