Le mer. 20 avr. 2022 à 18:47, Sigurd Kristian Brinch <sigurd.k.brinch@uia.no> a écrit :
Hi,
As far as I can tell, libvirt/KVM supports multiple vGPUs per VM
( https://docs.nvidia.com/grid/14.0/grid-vgpu-release-notes-generic-linux-kvm/...),
but in OpenStack/Nova it is limited to one vGPU per VM
( https://docs.openstack.org/nova/latest/admin/virtual-gpu.html#configure-a-fl... )
Is there a reason for this limit?
What would be needed to enable multiple vGPUs in Nova?
If you look at the vGPU types that are supported for multiple vGPUs per VM, those are only the ones who associate the whole GPU to one single vGPU (eg. A100-40C for the A100 40GB PCIe card)... You can try to ask for more vGPUs per instance if you want, but unless you use the above types (which are just kind of passthrough), you'll get the libvirt exception that's provided in https://bugs.launchpad.net/nova/+bug/1758086 ) This is then not a Nova limitation, but we tried to document in our upstream docs to let operators know about such limitation. -Sylvain
BR
Sigurd