What is the proper way to downgrade the nvidia drivers?

I am facing severe screen flickering with the current nvidia drivers (535). I want to revert back to the 520 one but I don’t know how.

I have tried the following:

sudo dnf downgrade akmod-nvidia

But after checking the driver version with nvidia-smi it is still the 535. I guess this happens because akmod-nvidia has already built a driver for my kernel(s).

Here is the output after running: dnf list installed "*kmod-nvidia*" (without downgrading):

Installed Packages
akmod-nvidia.x86_64                                                                                                                        3:535.113.01-1.fc37                                                                                                  @rpmfusion-nonfree-updates
kmod-nvidia-6.4.15-100.fc37.x86_64.x86_64                                                                                                  3:535.113.01-1.fc37                                                                                                  @@commandline             
kmod-nvidia-6.5.5-100.fc37.x86_64.x86_64                                                                                                   3:535.113.01-1.fc37                                                                                                  @@commandline             
kmod-nvidia-6.5.6-100.fc37.x86_64.x86_64                                                                                                   3:535.113.01-1.fc37                                                                                                  @@commandline 

Notice that kmod-nvidia* are installed from@commandline.

So based on this, I think I should remove all of them and then install the 520 version of akmod-nvidia. Is this correct?

PS: Why when I type dnf info kmod-nvidia I get:

Last metadata expiration check: 4:51:36 ago on Sun 15 Oct 2023 01:19:42 PM EEST.
Available Packages
Name         : kmod-nvidia
Epoch        : 3
Version      : 535.113.01
Release      : 1.fc37
Architecture : x86_64
Size         : 11 k
Source       : nvidia-kmod-535.113.01-1.fc37.src.rpm
Repository   : rpmfusion-nonfree-updates
Summary      : Metapackage which tracks in nvidia kernel module for newest kernel
URL          : https://www.nvidia.com/
License      : Redistributable, no modification permitted
Description  : This is a meta-package without payload which sole purpose is to require the
             : nvidia kernel module(s) for the newest kernel,
             : to make sure you get it together with a new kernel.

Why it doesn’t show the 3 installed packages?

This shows the kmod-nvidia package from rpmfusion and does not show the locally built packages. The locally installed packages are named kmod-nvidia-$(uname -r) when built and when you look at the installed names you posted you can see that in the package name. (example kmod-nvidia-6.5.6-100.fc37.x86_64)

The repos for rpmfusion and fedora only keep the most recent available packages (newest version) for installations. There are historical packages kept elsewhere, but I am not familiar with the location for those packages at rpmfusion.

Removal of all currently installed nvidia packages can be done cleanly withsudo dnf remove \*nvidia\* --exclude nvidia-gpu-firmware but a new install will only bring back the latest versions unless you locate and download all the needed packages with the desired versions and do an install from the local folder.

In my experience the only cards that need an older driver are those that are old enough to run only with the 470xx or 390xx drivers, and both those are available on the rpmfusion-nonfree repo.

The fact you are running on F37 (which is scheduled to become EOL in about a month) indicates the problem may be eliminated by upgrading to F38 which will have at least 6 more months support. Very few will be running the same version of fedora and even fewer will have your same video card so testing in an attempt to repeat your problems noted will be difficult.

Which version of nvidia-gpu-firmware is installed ? – it should show as 20230919-1 or newer.

What nvidia card is installed? Post the output of inxi -Gxxx so we can see that.

Thanks for the answer!

I will considering cleaning all nvidia packages (except the nvidia-gpu-firmware and akmod-nvidia) and then downgrade akmod-nvidia and see if it fixes the problem. It is really annoying since it has appeared from June 2023.

Yeah I am considering an update. I just don’t know if I will mess up anything with dnf system-upgrade :hot_face: since I am a little busy right now to perform a clean install.

In most cases that process works very well. The first step (download) always reminds you that you should do a full dnf upgrade --refresh step before beginning the download. I strongly recommend that you do so.

In several years of upgrading with that process I have never seen a problem occur.

Cleaning out nvidia is best done with the command I showed above and it also removes akmod-nvidia. You would need to remove akmod-nvidia anyway before it could be downgraded.

Does the problem still occur when you boot with the 6.4.15 kernel which seems to still be installed?

Remember that we can make no more than general guidance while you fail to provide requested information.

Thanks for the guidelines! I will test it on my laptop F36 (where I don’t care if I screw anything :sweat_smile:) and then to the desktop.

I don’t think the problem is related to the kernel because in so many upgrades (even in 6.4.15) it hasn’t gone away. Note, today I booted into windows and it is also happening there.