Frigate LXC with nvidia driver support

[xijio]

I briefly had the @remz1337 fork of the frigate script working with proxmox using NVIDIA-Linux-x86_64-550.135.run on the host using "bash -c "$(wget -qLO - https://github.com/remz1337/Proxmox/raw/remz/ct/frigate.sh)"". But then I was having trouble with the nvidia driver in another lxc and removed the 550.135 drivers and installed the proxmox nvidia-driver from apt (535.183.01). This broke the @remz1337 installed Frigate due to driver mismatch. I've been trying for a full day now to get either the lxc to match the driver of the host, or move the host back to 550.135 drivers and I'm faceplanting hard.

Passing GPUs to LXCs just seems really complicated. The @remz1337 fork worked for a while, but broke when I made a change and now when I try to run again to redeploy it errors out. I was going to post in that repo, but it has been archived as of the 17th. This brings me to my question:

Is there a possiblity of getting a frigate lxc script with nvidia driver support to be a part of the repo? If not, what is the recommended way to enable nvidia graphic support on the current script?

Thank you in advance!

[remz1337]

hey there, it's me. The original repo (tteck) was archived and so is now my fork of it too. You can still find my frigate script with gpu support on my new fork of the community-scripts repo (https://github.com/remz1337/ProxmoxVE/).

Indeed, for gpu passthrough to work in any LXC, drivers need to match exactly the host.

I have a script to install drivers across host and LXC:
bash -c "$(wget -qLO - https://github.com/remz1337/ProxmoxVE/raw/remz/misc/nvidia-drivers-host.sh)"

If you want an older version, download the file to your host and change this line 68 for the version you want to install (make sure it is listed here: https://download.nvidia.com/XFree86/Linux-x86_64/)

[xijio]

Hi Remz, I totally missed this over Christmas! Thanks a ton for replying to me, I really appreciate it! I think your nvidia driver script is really cool and might be considered as an install option in a bunch of different LXCs in the future. As we move more and more toward AI GPUS are getting more and more popular in selfhosted applications and I think it'd be great to be able to easily manage the drivers on the host and in each lxc with community scripts. For now though I'll definitely take a look at the nvidia script. Thanks a ton!