extensions Update NVIDIA drivers up to version 550.x

Hello.

Current NVIDIA drivers in Talos are quite outdated. 535.129.03 version has been released six months ago. This version has CUDA 12.2, which is a problem for some new software, such as ffmpeg v7.

Is there any ETA for upgrading it to the latest 550.x branch?

Jun 11 '24 11:06 maxpain

There is no ETA at the moment, but update will happen at some point before the next Talos release.

Jun 11 '24 11:06 smira

Also updating to latest nvidia seems to break enterprise users, so we follow caution in updating nvidia drivers

Jun 11 '24 14:06 frezbo

Also updating to latest nvidia seems to break enterprise users, so we follow caution in updating nvidia drivers

I think with this update we should also support older drivers (need to figure out extension names vs. Talos upgrades), but basically provide both 535 and 550 for Talos 1.8.

Jun 11 '24 14:06 smira

Yes, we should figure this out

Jun 11 '24 14:06 frezbo

Theoretically, can we add some option to choose the extension version in factory.talos.dev?

Jun 12 '24 10:06 maxpain

No, you can use imager if you want your own custom image. And for NVIDIA it's anyways locked to the kernel build, so it won't work for a different version of Talos

Jun 12 '24 11:06 smira

If you get any hot build, I can test it... I would love to have an updated version. I've tried to play with the imager and some modules from here taking the same kernel version from nightlies, but it didn't boot up :(

Aug 18 '24 20:08 daper

I have a workflow to build the kernel and the Nvidia open kernel modules here, if you need inspiration. I'm running 6.10.3 with NVIDIA 550.107.02 on my nodes.

Aug 18 '24 20:08 jfroy

Starting Talos 1.8, both LTS and Production NVIDIA versions will be shipped as per support matrix: https://www.talos.dev/v1.8/introduction/support-matrix/

Aug 19 '24 02:08 frezbo