545.23.08 is not listed in the NVIDIA Driver Downloads Page

As mentioned in the title, 545.23.08 is not listed here: Official Advanced Driver Search | NVIDIA.
The latest version available there is 545.23.06.
The CUDA Linux repository, and the downloads page for CUDA are serving 545.23.08 however.
Unfortunately, in Unix Drivers | NVIDIA as well, the latest Beta version is listed as 545.23.06 and not 545.23.08.

3 Likes

Hi @sagnik thanks for bringing this up.

With some releases it can happen that the CUDA packaged driver is slightly ahead of the downloadable driver. Being a patch version differenc it should not make a difference.

Nevertheless I will let the CUDA team know that this mismatch exists.

Thanks!

3 Likes

Thank you. I’m primarily facing issues because flatpak only has a package for .06, so all flatpak apps are no longer GPU accelerated as a result.

2 Likes

Hi @sagnik, would you mind clarifying this a bit? How does this cause issues?

If latpack apps get installed with the .06 version then CUDA will still work with that version. And if CUDA installation overwrites the .06 driver with .08, all apps also should still work if they simpy use CUDA.

Do you have some kind of error log or similar?

I am not exactly sure how the mechanism works, but it appears that flatpak packages the Nvidia driver themselves and that is used for running flatpak apps with GPU acceleration.
At the moment, the closest matching version in flatpak is .06 and my system is running on .08, so I believe that is why no flatpak packages for the Nvidia driver are being downloaded, and hence no GPU acceleration for flatpak apps.

Here is the build script on their GitHub that defines the versions that the package will be built for: https://github.com/flathub/org.freedesktop.Platform.GL.nvidia/blob/master/versions.sh .

Just to clarify, this is not a CUDA specific problem, but rather a problem with the driver version supplied by the CUDA Nvidia repository, and the mismatch with the version listed and available publicly (through .run files)

Thank you for the clarification!

That is really unfortunate.

Well, we do try to avoid these situations where regular and CUDA driver diverge, this should only happen in cases of hotfixes or similar. I hope this will align itself again soon!

Having this problem here too. I need to pull the 545-29-08 as half of my GPU accellerated flatpak apps no longer work anymore.
Would you recommend I go through the (highly error prone) process of downgrading my driver to 545-29-06 or should I expect some sort of alignment in the near future. None of my failed flatpak aps are mission critical, just been annoying not having them available on my system.

Thanks.

Hi @dannyhepple and thank you for joining the NVIDIA developer forums.

Right now, if flatpak is not essential, I would be patient and wait for the next driver update, which I am sure will come soon. Then version numbers should be in line with the general driver train.

Thanks!

Thanks for the reply Marcus. I think I’ll just do that.
I have a question for you to if you’ve got the time relating to this issue.

So I’m currently on a RHEL 9 system and I’ve installed the 545.23.08 driver directly from the Nvidia repository using sudo dnf config-manager --add-repo http://developer.download.nvidia.com/compute/cuda/repos/rhel9/$(uname -i)/cuda-rhel9.repo
and then sudo dnf module install nvidia-driver:latest-dkms

By using this method have I installed a certain CUDA variant of the driver preferred for development purposes or is this just the normal driver I’d install if I was using the system for 3D content creation and graphics rendering? Is this the exact driver that would make its way into other repos when it’s released?

The reason I ask is that I’ve noticed I can install the drivers from repositories like RPM fusion too. I’m just wondering if I was to install the driver from a different source, perhaps I’d have less chance of misalignment in the future? Just trying to make sure I’m not installing some alternate CUDA version of the driver I don’t actually need.

We run a lab for students with many workstations for CGI. Our main apps are custom deployed, but many additional apps are flatpaks and it’s very inconvenient to run into this situation after an upgrade.

Do you have a timeframe, when the versions will be aligned again ?

Hi @jeanf, welcome to the NVIDIA developer forums.

This is hard to say, the most current Windows driver version has reached 546, so I expect a 546 version to be released for Linux as well pretty soon. But I do not have an exact ETA, sorry.

any update for this issue?

Hi @tom266, welcome to the NVIDIA developer forums.

If you check the driver download page and select “New Feature branch” you will see v545.29.06. So if flatpak updates with the latest of the latest, which caused the issue in the first place, things should be fine again now.

im running cuda and lxc on my host, adding nvidia permissions via cgroups, and manually installing guest drivers with the NVIDIA-Linux-x86_64-${version}.run file with the --no-kernel-module flag ala Nvidia GPU passthrough in LXC :: TheOrangeOne

i had to downgrade to cuda_12.3.0