I have two proxmox nodes that have Nvidia Tesla P4s in them. I have been stuck having to pin kernel version 6.5 on these nodes for awhile now due to the vGPU kernel modules not building successfully for newer kernels.
Today I noticed when updating the official vGPU host drivers to version 16.6 that the vGPU kernel modules build successfully for the current 6.8.8-2-pve kernel. I booted one of these nodes into this kernel and am seeing the following:
1. nvidia-smi shows the expected output.
2. nvidia-smi vgpu shows the expected output.
3. mdevctl list returns an empty line.
4. mdevctl types returns an empty line.
It seems like the drivers are almost working. Everything is working correctly on kernel version 6.5.13-5-pve on both of these nodes, so I know everything is configured correctly for that kernel version. I already have the proxmox-default-headers package installed on both nodes, as well as the proxmox-headers-6.8.8-2-pve package. I checked the logs from the nvidia installer and am not seeing any errors. I am running the stock vGPU drivers from Nvidia, so not the modified unlocked drivers.
Are there extra steps required to get mdevs working for vGPU on this newer kernel version? Or are the official 16.6 vGPU drivers still not compatible with the latest pve kernel?
Today I noticed when updating the official vGPU host drivers to version 16.6 that the vGPU kernel modules build successfully for the current 6.8.8-2-pve kernel. I booted one of these nodes into this kernel and am seeing the following:
1. nvidia-smi shows the expected output.
2. nvidia-smi vgpu shows the expected output.
3. mdevctl list returns an empty line.
4. mdevctl types returns an empty line.
It seems like the drivers are almost working. Everything is working correctly on kernel version 6.5.13-5-pve on both of these nodes, so I know everything is configured correctly for that kernel version. I already have the proxmox-default-headers package installed on both nodes, as well as the proxmox-headers-6.8.8-2-pve package. I checked the logs from the nvidia installer and am not seeing any errors. I am running the stock vGPU drivers from Nvidia, so not the modified unlocked drivers.
Are there extra steps required to get mdevs working for vGPU on this newer kernel version? Or are the official 16.6 vGPU drivers still not compatible with the latest pve kernel?