Hello,
I’m permit to open a topic to know if it’s possible to run vGPU Manager on flatcar, along with kubernetes and kubevirt.
For the context : I’m currently testing kubevirt on flatcar bare metal servers. Working well for now. I can manage VM inside kubevirt.
But well the graphics part of KVM is a little too slow for virtual desktop. So we would like to try vGPU with Nvidia A2 cards to accelerate the remote display rendering.
I manage to create an offline (air gapped) container with the nvidia driver for flatcar. Kernel modules are loading, nvidia-smi (inside the container) works ( return informations about the physical Nvidia A2 card).
But, If I understand correctly the documentation and the Helm chart of GPU Operation, I need to build a container image “vgpu Manager”, right ?
And I don’t see any dockerfile for flatcar. Is there any possibilities by the way ?
Or impossible ? (I’m not asking for official support of course, just the possibility to run and try it)
For the driver container I use :
the Flatcar Dockerfile from the nvida gitlab repod with driver version 510.108.03 (support of cuda 11.7.1 as use in the Dockerfile and support nvidia A2 card as mentionned in grid release compat pages.
But for vGPU manager only dockerfile for Ubuntu ? no Flatcar
Perhaps with writing my own Dockerfile ? Or impossible ?
Thank you in advance
Please feel free to ask for details or others informations.
Regards,