Enabling Alderlake Quicksync in VM and Applications

[tsaG]

Hey!

I enabled the Quicksync/VGA Driver on my Alderlake CPU and I can find it in /dev/dri as renderD128 as it should be.

For apps, I can select how many (?) Quicksync devices should be connected to the Container.

Is this correct? Is this really the amount or should it be the specific Device? I mean it would make sense both ways. It could also be Device 0 to 5 (if that would make any sense).



I selected 2 of them to my Jellyfin container (I still have to configure jellyfin to make it work, thats a problem for future me).

However, I would like to attach (at least one) Quicksync device to my VM for Home-Asssistant / Frigate.

How would I see the correct PCI ID to do so?

This is the output of my lspci doesnt give me the Information "PCI_" I am looking for.

 

[LarsR]

You can use quicksnyc with up to 5 container at the same time, but the container cant use the same gpu id so there are 5 different id's for 5 container

 

[truecharts]

You can use quicksnyc with up to 5 container at the same time, but the container cant use the same gpu id so there are 5 different id's for 5 container

No.
It means "Allocate 1 GPU of said type" (or more) nothing more, nothing less.

 

[tsaG]

No.
It means "Allocate 1 GPU of said type" (or more) nothing more, nothing less.

Okay, as I only have one CPU and thus also GPU I would assume its the quicksync Cores. So I can assign 4 "Quicksync" Cores to the Apps and the other 1 to my VM?

Do you know how I would get the PCI ID of the GPUs to add them to a VM? In the VM Section I can not add those Quicksync cores but would have to add them as a PCI device "pci_00..." (I guess).

 

[truecharts]

Okay, as I only have one CPU and thus also GPU I would assume its the quicksync Cores. So I can assign 4 "Quicksync" Cores to the Apps and the other 1 to my VM?

Do you know how I would get the PCI ID of the GPUs to add them to a VM? In the VM Section I can not add those Quicksync cores but would have to add them as a PCI device "pci_00..." (I guess).


View attachment 58088

No it just means allocate 1 GPU to the App. Like stated: nothing more, nothing less. Not cores, not something special. Just what was explained.
You need to ignore the rest.

You cannot assign a GPU to both Apps and VM's however.

 

[rs_taylor]

What is the significance of the 5 GPU then, just a bug?

 

[truecharts]

What is the significance of the 5 GPU then, just a bug?

It requests (allocates) in a multi-node multi-gpu cluster. It's really not of any concern of end-users.
It's basically a fluke in how Intel only allows multi-pod use of GPU's if you split the GPU into multiple virtual gpu's.