NVIDIA-Driver-510.73.05 - Failed to build Modules

My current version 510.68.02 is working fine.

Trying to update to the latest proprietary NVIDIA drivers (The proper way) by building the .run file in “systemctl set-default multi-user.target” mode…

I know there is the precompiled cuda-rhel8.repo route, but I’ve had borked system on reboot issues from that…

Historically, I’ve found the most stable way to be to manually build in multi-user.target…

Maybe the new backported 8.6 kernel is the issue? I’m kind of afraid to try again until there is a newer driver…

Anyone else experiencing this? Can’t wait for open-sourced kmods to take effect!

I’ve always used “third way”:

dnf install elrepo-release
dnf install kmod-nvidia

@jlehtone thank-you, that looks about the most concise way to do it!
Any experience with Maya or Davinci Resolve?
I know the rpmfusion repo version is hit-or-miss with CUDA stuff…

Is this elrepo-release/kmod-nvidia basically a pre-compiled version of the latest NVIDIA .run file?

Cheers!

@jlehtone Actually, this is a very outdated version… I need absolute latest for Redshift Render

…Or is my rpmfusion repo overriding?

image

Those are clearly packages from rpmfusion-nonfree-updates.

The NVidia .run file is just a tarball or whatever with a shell script. The package management has no idea what files it places, or worse replaces.

NVidia have their own yum-repository, where they have both the graphics driver and CUDA as RPM packages.
RPMFusion (a yum-repository) has the graphics drivers as RPM packages.
ELRepo (a yum-repository) has the graphics drivers as RPM packages.

RPM-files the package management can install/update/remove “cleanly”.

ELRepo’s additional twist is that their kernel modules work with all kernels of one point update. The others have you install or build the kernel module for each kernel.

Yes, ELRepo has currently only version 470.103.01-2.el8_6 for EL8, while they have 510.68.02-1.el7_9 for EL7. If you want latest and particularly if you need CUDA, then NVidia’s yum repo is the best bet.
See Installation Guide Linux :: CUDA Toolkit Documentation

The elrepo-release is an RPM package from Rocky’s ‘extras’ repository that provides files that define the ‘elrepo’ repository. You can see the files of RPM package with:

rpm -ql elrepo-release    # if package is installed
rpm -qlp elrepo-release-8.2-1.el8.noarch.rpm  # if you have that RPM-file
dnf rq -l elrepo-release  # for package that is in repo, but is not installed

Anyone else fail to build kmods from the latest 510.73.05 NVIDIA update (from NVIDIA .run file)?

Ryzen 9 5900x + three 3080ti’s

I’m interested to know more about this. Are you saying they have signed kernel modules that work with multiple kernel versions?

Yes. Normally when you compile a module, it works only with the kernel that you did build it for. RHEL kernels within one point update are so similar that if you strip the exect version requirement from the module, it will work with them all. For example, all 8.5’s kernels could have been used with one ELRepo module. The 8.6’s kernel is different and ELRepo has rebuilt the kmod-nvidia to match.

And yes, ELRepo has their own key to sign their modules with. One can import that key into UEFI and then one can have Secure Boot enabled.

Thanks, this is really interesting; do you how they strip the version; do they use something like an ‘elf’ editor on the final module.