RmInitAdapter failed on Fedora 43 for RTX 5060 Ti

Graphics / Linux Linux
,
1

nvidia-bug-report.log.gz (628.5 KB)

Summary: Purchased new card intended for use on my Linux Fedora desktop. Recently upgraded to Fedora 43 which released last month. Installed recommended driver. After reboot, nvidia-smi didn’t find a device, even though the nvidia kernel modules had loaded. Journalctl showed rm_init_adapter failed. Tried two alternative driver versions, no luck.

Card: Gigabyte GeForce RTX 5060 Ti Windforce OC 16GB rev 1.0
OS: Fedora Linux 43 (Cinnamon spin) Kernel 6.17.7-300.fc43.x86_64 on x86_64

_______________________________________________________
| Driver      | Result                               |
| 580.105.08  | nvidia-smi didn't find device        |
| 580.82.09   | nvidia-smi didn't find device        |
| 575.64.05   | driver installation errored and quit |
------------------------------------------------------

Had to go into bios to set PCIeX16 bifurcation to 8x8x before the card worked.

580.105.08 was the recommended and latest driver for linux for my card. I reinstalled it 3 times and tried multiple reboots. nvidia kernel modules loaded, nouveau was not loaded. Journalctl showed these messages: “NVRM: RmInitAdapter failed! NVRM: rm_init_adapter failed, device minor number 0”. Lspci showed the card.

Dual boot to Windows 10 on the same hardware (even diff partition on same hdd) and the card works with the recommended Windows driver.

Questions:

  • Is the problem that my card is too new and hasn’t been added to driver support?
  • Is the problem that Fedora 43 is too new? Would it work if I backtracked to Fedora 42? There’s no reason I have to be on 43.
2

I think the source used for the Fedora (rpmfusion) NVIDIA drivers is pulled from here and you might look at the README to see if your GPU is listed in Appendix A. Supported NVIDIA GPU Products.

Run the following which should show the exact device id for you GPU:

lspci -vnn | grep -i 'vga.*nvidia'
01:00.0 VGA compatible controller [0300]: NVIDIA Corporation GA104M [GeForce RTX 3080 Mobile / Max-Q 8GB/16GB] [10de:249c] (rev a1) (prog-if 00 [VGA controller])

In my case the vendor id is 10de (NVIDIA) and the device id is 249c. Then go to the above link for the Supported Products and search for your GPU by name (RTX 5060 Ti) and see if the device id listed (2D04) matches yours.

I can’t say for certain that means your specific card is supported, but it seems like a reasonable assumption.

As to F42 vs F43… since I think the same source is used to build the drivers for both F42 and F43, I suspect switching to F42 wouldn’t help. But that’s really just speculation on my part.

If you have space on your drive to create another partition and install F42, that might be a good test. Maybe you could shrink your F43 partition and add a new one just to test F42. If there’s no diff, it should be easy to delete the test F42 partition and expand the F43 back into it. Just guessing, but I suspect you could install F42 with a15G partition, maybe even 10G. If nothing else, this is a great way (IMO) to have a test partition where you can fiddle with different setups without fear of messing with the OS you actually use every day.

3

Thank you for your comment. I checked the device ID and it is indeed 2d04, which is listed as supported.

I think you’re probably right that f42 and f43 doesn’t matter, and am currently troubleshooting based on that assumption. As such I haven’t tried your idea of f42 on a new partition yet, but I think it’s a good one and I may end up trying it later.

I’ve been installing nvidia drivers on fedora for years following this guide:
https://www.if-not-true-then-false.com/2015/fedora-nvidia-guide/

However, this morning on these forums I found an official nvidia guide:

So I just ran the uninstall script and tried that. Unfortunately it seems quite impossible to follow this install guide, as without a browsable list I just get 404 “not found” at every turn. For example,

https://developer.download.nvidia.com/compute/nvidia-driver/580.105.08/local_installers/nvidia-driver-local-repo-fedora42.580.105.08.x86_64.rpm

Returns 404. This guide seems to be setting users up to fail. Why is there no link to a resource where you can see what’s available? Is it there and I’m blind to it somehow?

Now I’m thinking I should try the guide at RPM Fusion:

Years ago this did not allow me to install drivers that were recent enough. But perhaps times have changed? I’m skeptical that it will work with my card since it’s new. But as far as things-to-try go, I guess it might be next.

If I can’t resolve soon, or at least before the retailer’s return window expires, I may have to return the card and revert back to my RTX 2070 Super 8GB.

4

I got an unexpected (by me) result when I tried the RPM Fusion guide–it may have… just worked out of the box:

nvidia-smi shows 5060 Ti, driver 580.105.08, cuda 13.0
I was able set-default graphical.target and boot into the GUI!

That would (of course) be ideal if everything’s just working now. I have to reinstall some software that was broken by the Fedora 42 => 43 upgrade, and then do some testing to be sure. I’ll aim to come back after that and mark this issue as resolved if it pans out.

Years ago RPM Fusion wasn’t working for me as an option for nvidia drivers. I’m glad I tried again today.

For now just wanted to add a quick note to say thanks to forbyta, anyone else who looked at this, and to the people responsible for RPM Fusion support.

If you’re having a problem, try their (easy) how-to-nvidia guide!

5

I was just about to post the following when I noticed you appear to have got it working, so the rest of this may be irrelevant. At any rate here’s my take on NVIDA Cude vs RPM Fusion. Perhaps there’s some tidbits useful for someone in the future.

NVIDIA Cuda Repository Approach

I also recently found the NVIDIA Driver Installation Guide and I found it a bit confusing with respect to getting the correct URL. I ended using the Network Repository Installation section instead of the Local Repository Installation.

If you look at the URL to enable the network repository - specifically the part before the variables, ie this bit, which you can browse to:

https://developer.download.nvidia.com/compute/cuda/repos/

Then from there, you can navigate down to find the full URL to use with the dnf command.

Unfortunately, you’ll discover “fedora43” is missing - I assume they just haven’t added it yet.

I’m pretty sure you can use the F42 repo on F43, but I could be wrong. I tried it on F43 and the drivers installed fine and they worked just as well as they did when installed on F42. That is to say they didn’t work, but they didn’t work in the same way (see here for the problem I’m trying to fix). So, I guess that’s not exactly a glowing endorsement for compatibility. Fact is, I haven’t really been able to thoroughly test any version of 580 because of the above issue.

At any rate, if you want to try to use the NVIDIA cuda repo, this is the dnf command and correct URL you’d use to add the F42 repo:

sudo dnf config-manager addrepo --from-repofile=https://developer.download.nvidia.com/compute/cuda/repos/fedora42/x86_64/cuda-fedora42.repo

Then from there, you just need to proceed with the Driver Installation part of that guide.

RPM Fusion Repository Approach

Now (IMO), specifically for Fedora, I think the RPM Fusion approach is easier. The trickiest part is just adding/enabling the correct repo.

For the official releases (KDE, Gnome) - I think the packages are per-installed, you just need to enable the repos. It’s been a bit since I installed either KDE or Gnome and I’ve been testing a bunch of different distro lately , so it’s all gotten a bit muddled. Assuming that’s true, you can run this to see which repos are currently enabled:

sudo dnf repolist

And to see which repos are available, run this

sudo dnf repolist --all

For F43 KDE Plasma (for example), you (may) need to enable the nvidia repo. ie, if it does not show up with “dnf repolist” but does show up with “dnf repolist --all”.

dnf config-manager setopt rpmfusion-nonfree-nvidia-driver.enabled=1

For F43 Xfce on the other hand, the packages are NOT pre-installed, so they need to be installed first based on the instructions from here. ie.

sudo dnf install https://mirrors.rpmfusion.org/free/fedora/rpmfusion-free-release-$(rpm -E %fedora).noarch.rpm https://mirrors.rpmfusion.org/nonfree/fedora/rpmfusion-nonfree-release-$(rpm -E %fedora).noarch.rpm

Then once the packages have been installed you should be able to see the disabled repos by running “dnf repolist --all”. And at that point you just need to enable the nvidia repos:

sudo dnf config-manager setopt rpmfusion-nonfree.enabled=1
sudo dnf config-manager setopt rpmfusion-nonfree-updates.enabled=1

NOTE: I’ve never used the Graphical Setup from the RPM Fusion Configuration page. It might enable the repos as well.

For F43 Cinnamon I’m afraid I don’t know - I’ve never installed it. Maybe it already has the repos (like KDE/Gnome), maybe they’re even enabled. Or maybe it’s like Xfce where the repos need to be added.

At any rate, once the correct repo is added/enabled, then you just need to install the drivers:

sudo dnf akmod-nvidia xorg-x11-drv-nvidia-cuda

The “xorg-x11-drv-nvidia-cuda” package is optional depending on your needs. I personally only install it because the “nvidia-smi” command comes with it and that’s a useful tool.

WARNING - the dnf command will exit after the packages have been installed, but the drivers themselves haven’t been built yet - that’s happening in the background. You’ll want to wait until that’s done before you reboot. I usually run top and wait until the various building commands have finished. You can also check the build log… I think it’s in /var/cache/akmods/nvidia.

Once it’s all done, reboot.

6

Thank you, that looks like it!

And your post is helpful, as I’m still trying to align software packages with respect to CUDA version, compiler version, etc. Also, as you say, it may help someone else too.

Yeah RPM Fusion worked great for me for the driver, and for adding basic CUDA support. I expect I’ll be using it again soon as I continue to work on my environment.

7

I can now confirm that I have a working linux driver for my card! This was the solution for my situation & system:

  1. Run the nvidia uninstall script found in my PATH to clean out my failed driver installation.
  2. Follow the RPM Fusion “How to nVidia” guide to install the drivers.

That was it! nvidia-smi showed the card, and last night I was able to install the software I needed to actually use it, and it worked!

Thank you!