I recently updated to f42 beta and was originally using Nvidia’s repository for GPU drivers since they were more up-to-date than what was available via RPM-fusion.
Everything was still working after the upgrade from F41, but I noticed some package clashing between the RPM-fusion version and Nvidia version of packages while doing another dnf update (oops, I should probably have disabled the RPM fusion ones), so I decided to uninstall all GPU drivers to be safe and reinstall the RPM-fusion version since it had the latest drivers anyway and Nvidia has no F42 repo yet.
So I uninstalled, rebooted, then followed the RPM fusion install instructions, including waiting until the version properly shows before rebooting…and from what I can tell, it seems to have successfully installed.
Now, the GPU driver does not seem to function, and it seems like the root cause is based on this in dmesg:
NVRM: The NVIDIA GPU installed in this system requires use of the NVIDIA open kernel modules.
I’m using a 5070 ti so the requirement for the open kernel module is not a surprise…I’m wondering if maybe the wrong kernel is getting chosen somehow.
I tried force rebuilding akmod, dracut, initramfs, nothing worked.
nvidia-smi also is not detecting hardware.
When installed the same version with the Nvidia .run file and explicitly choose the open kernel, everything works as expected (I had to self sign it since I use secure boot).
I’m not sure whether to report this here or to Nvidia somewhere. Please let me know what other details would be useful. I hope it’s just me and something i can fix on my end with some help. Thanks!
Thanks, I apparently missed this section (or really, didn’t know to look at that section). I’ll give that a try and report back!
This might be a common enough issue with newer gpus that it might be worth mentioning it in the main install section so it’s not so easily missed, but I understand why it’s compartmentalized as it is now.
It has been mentioned many times in recent threads that the newest nvidia GPUs (the 5k series) do not work with the proprietary drivers on linux and that only the nvidia open drivers are used for those cards.
People who frequent this forum should be aware that nvidia is migrating from the proprietary drivers for linux which work for all the 1K to 4k series cards to the open drivers which work for the 3k+ series cards (I don’t use a 2k series card so have no experience with the open drivers on that series.)
My apologies, I am admittedly not a (direct) frequenter of this forum.. I just wanted to contribute feedback since I find myself using Fedora more, and in case this was an unknown bug and at least felt I could adequately report it. I wasn’t finding any (recent) direct hits for the error message, so I thought I’d post.
Because of previous searches I’ve made (probably from hits here, as you mentioned) for other reasons, I was at least aware of the open kernel requirement for the 50 series, which I feel is how I was able to understand the issue at all.
That said, I don’t think it’s an easy thing to discover, much less understand, what’s wrong in this case; the install appearing to succeed makes it seem like some sort of setting with grub or dracut or whatever was not correctly incorporated, not that the entirely wrong kernel was installed.
It’s very easy to go down the wrong rabbit holes with it, and I feel it might be a leap to assume most 50 series users will be privy to this issue or recent forum activity, much less understand how to investigate it or figure out how to alleviate it.
I totally understand that this is a WIP and I’m happy to use an easy workaround, but I definitely think it would be beneficial to at least add a 1 line sentence to the install section referencing this necessary workaround for 50 series which links to the Kernel Open section, rather on relying on the user to understand this innate requirement/relationship.