documentation/docs/admin-guide/troubleshooting/nvidia.md

Nvidia

Failing to Install Nvidia Driver

Some systems may fail to install the correct Nvidia driver even after several reboots. While running puppet you usually then see following error:

Notice: /Stage[main]/Profile::Nvidia::Cuda/Exec[Nvidia_check_failed]/returns: failed to install/loaded Nvidia software configured by Puppet as there is still the Nouveau driver loaded, please reboot

To get more details on the problem you can run the following command:

bash -x /opt/pli/libexec/ensure-nvidia-software

If you see the following error message:

+ dnf --quiet --assumeyes module install nvidia-driver:580-dkms/default
Error: Transaction test error:
  file /usr/share/egl/egl_external_platform.d/15_nvidia_gbm.json conflicts between attempted installs of egl-wayland-1.1.13.1-3.el9_8.x86_64 and egl-gbm-2:1.1.3-1.el9.x86_64

the issue can be resolved by following the steps below:

1. Install the egl-wayland package:

   dnf install egl-wayland
   

2. Re-Execute the failing command (omit the --quiet and --assumeyes flags to see more details) - this command will fail again with a different error message - you can ignore this error message and continue to the next step:

   dnf module install nvidia-driver:580-dkms/default
   

3. Find rpm

   find /var/cache/dnf -name "*egl-gbm*"
   

4. Install package from the rpm file found in the previous step:

   rpm -i --force $(whatever you found in the previous step)
   

5. Again re-execute the failing command (this should succeed now):

   dnf module install nvidia-driver:580-dkms/default
   

6. Reboot the system and re-run puppet to ensure that the Nvidia driver is installed correctly.