[SOLVED] Proxmox VE 7.0 CUDA Problemlösung

[Sub7]

Es soll ja bekannt sein dass CUDA mit Debian Kernell 5.9/Proxmox VE 7.0 inkompatibel ist.
Wie viele andere hätte ich beim ausführen von "pve6to7" eine Warnung gewünscht.

Lösung (LXC):

HOST & LXC Uninstallieren NVIDIA Treiber

sudo apt-get remove --purge '^nvidia-.*'
or/and
sudo /usr/bin/nvidia-uninstall

HOST

wget https://us.download.nvidia.com/XFree86/Linux-x86_64/460.73.01/NVIDIA-Linux-x86_64-460.73.01.run
chmod +x NVIDIA-Linux-x86_64-460.73.01.run
./NVIDIA-Linux-x86_64-460.73.01.run

LXC/GAST

wget https://us.download.nvidia.com/XFree86/Linux-x86_64/460.73.01/NVIDIA-Linux-x86_64-460.73.01.run
chmod +x NVIDIA-Linux-x86_64-460.73.01.run
./NVIDIA-Linux-x86_64-460.73.01.run --no-kernel-module

...mein Projekt wo ich CUDA nutze -> https://github.com/Sub-7/FFmpegUI

 

[bgv2020]

So you were able to fix by removing the nvidia drivers first, then installing afterwards?

No changes to your configuration file for cgroups to the new cgroups2?

 

[Sub7]

From version 7 onwards, changing the group (/etc/pve/nodes/PVE/lxc/100.conf) is mandatory, otherwise pass-through does not work.
Change:
"lxc.cgroup.devices.allow: c 226: * rwm"
to
"lxc.cgroup2.devices.allow: c 226: * rwm"

Now reinstall the correct Nvidia driver and it will work (see above).
PS: Host and guest need the same driver version.

 

[bgv2020]

Thank you!

 

[bgv2020]

This is a bit off topic but I'm nervous about doing a distro-upgrade and not being able to be successful doing the LXC Nvidia passthrough. Is it possible to do a snapshot of rpool (zfs) and then do the upgrade. If the upgrade doesn't work out then rolling back to the snapshot taken?

 

[Sub7]

My personal experience with upgrade:
2x Buckup HDDs were completely deleted, I only found empty folders.
... maybe I missed something.

My advice: make / copy a backup to an external HDD !!!​

After the upgrade, all containers were there, groups changed, drivers renewed, and everything works again.
i used that => https://www.youtube.com/watch?v=RCSp6gT7LWs

 

[bgv2020]

My personal experience with upgrade:
2x Buckup HDDs were completely deleted, I only found empty folders.
... maybe I missed something.

My advice: make / copy a backup to an external HDD !!!​

After the upgrade, all containers were there, groups changed, drivers renewed, and everything works again.
i used that => https://www.youtube.com/watch?v=RCSp6gT7LWs

Thank you so very much for your help. I will attempt this weekend!