I installed this on another similar system and did not have an issue.
I am running RHEL 8.1 on a power9 system.
I am attempting to install the 11.1 cuda toolkit with nvidia-drivers.
I have disabled nouveau and run dracut --force
I followed the instructions here without error:
What I have noticed is that the command modprobe nvidia fails with driver not found in the /lib/modules directory.
On the system that is working I find that there are bunch of nvidia drivers under /lib/modules/*/extra/ that are not present on the system that is not working.
I have the kernel-headers and kernel-devel installed.
What am I missing for those files to be created?
Suggestions?
Ok I figured it out and its bizarre.
I had to install the IBM xlc toolset which shoved in a bunch of perl dependencies, then uninstall nvidia-driver.
I then run dnf module install nvidia-driver:latest-dkms
it suddenly built the /usr/modules/(kernel-version)/extra files.
There was even extra files it installed on the dnf, it was installing 62 files before xl was installed, but after xl install it was installing 83 then it actually ran this step:
running scriplet: kmod-latest-dkms-3
So to be clear both times I ran the exact same commands:
dnf module install nvidia-driver:latest-dkms
So then to verify it was XL and not the local install of the cuda repo I did the install on the other IBM system I had just running the install from the repo. It did not run the scriplet and it was back to 62 files installed.
Now when I installed xlc I had the export PATH and LIBRARY_PATH set to include the cuda files, even though the drivers did not work. I don’t know if its necessary to install nvidia-driver, install xlc, and then reinstall nvidia-driver but that is what I did.
That is very strange.
I never got an error in any of the installs.