[Resolved] Discover and DNF state different versions of akmod-nvidia

Ask Fedora
, , ,
1

After performing a pass of akmod force rebuild, the problem has been solved.

  • GPU undetectable: driver version mismatch than kernel object
  • Works without nvidia-gpu-driver: old version has already baked into kernel object

Now the driver is @595.71.05 with working framebuffer.

I’m using KDE Plasma Desktop Edition for my main system’s Operating System;
Using

  • Intel i5-13400F CPU;
  • NVIDIA RTX 3070 GPU;
  • and 32GB Phy. Memory.

I recently upgraded my system to F44, and GPU driver broke.
After inspecting some discussions incl. Ask Fedora, I decided it’s because of the version confliction, or at least the defective version on my system.

So I purge-reinstalled all of them:

sudo dnf remove "*nvidia*" -y
sudo dnf install akmod-nvidia xorg-x11-drv-nvidia-cuda nvidia-container-toolkit
sleep 300 && sudo nvidia-ctk runtime configure --runtime=docker
sudo reboot

Note: Wayland will hang when you perform this since nouveau will crash. If it does, give your system a minute and just spam Ctrl+Alt+Del to force restart. Kernel is still alive, the framebuffer is the only one hung.

Using this, RPM Fusion provides 595.58.03 version of akmod-nvidia and it just works.

However, when I checked Discover, it says the driver should be upgraded into 595.71.05. When installed, the entire system doesn’t recognize my GPU anymore. (I tried using that specific version also.)

Version differs bwn RPM Fusion and Discover
Version differs bwn RPM Fusion and Discover645×427 32.3 KB

Idk why this is happening, but one thing is sure: 595.71.05 doesn’t work on my system.

I wanna know if this issue is reproducable on others’ system. On my system, the rate is 100%.

Suspecting driver bug, I opened Journald and found this:

Part of Journal under kmod-nvidia 595.71.05
Part of Journal under kmod-nvidia 595.71.051268×270 111 KB

The driver has falled back into nouveau (means 595.71.05 doesn’t work on my system), and nvidia-cdi-refresh has crashed with status FAILURE.
Using 595.58.03, nvidia-cdi-refresh will run as intended and finishes successfully.

Reinstalled nvidia-gpu-firmware:

sudo dnf install nvidia-gpu-firmware

Changes undetectable.

2

I haven’t had any issues

 nvidia-smi
Thu May  7 00:36:18 2026       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 595.71.05              Driver Version: 595.71.05      CUDA Version: 13.2     |
+-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA GeForce RTX 3050        Off |   00000000:01:00.0 Off |                  N/A |
|  0%   36C    P8              5W /   70W |     212MiB /   6144MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A            1629      G   /usr/libexec/Xorg                       169MiB |
+-----------------------------------------------------------------------------------------+
3

Who told you running this was a good idea?

sudo dnf remove "*nvidia*" -y

You deleted everything with nvidia in the name.
Of course the GPU doesn’t work anymore, you deleted more than just the drivers.
You erased nvidia-gpu-firmware. The kernel doesn’t even know how to load drivers for nvidia gpus anymore.

Put it back.

4

It worked for me at least, but if it was a bad idea, it must be my fault.
Now have I installed it and requested to reboot; All components work as before.

5

Quick note: above screenshot of nvidia-smi has been captured w/o nvidia-gpu-firmware. Before-After changes are undetectable for me.

6

It would only be noticeable after a reboot.
Install the firmware and proprietary drivers again, then manually run akmods using the following code block

sudo dnf remove akmod-nvidia xorg-x11-drv-nvidia
sudo dnf install nvidia-gpu-firmware akmod-nvidia
sudo akmods --force --rebuild
  • If it suceeds and just says [OK], then reboot and it should be working.
  • If it errors, you can share the driver build log with following command:
    fpaste /var/cache/akmods/nvidia/.last.log
7

akmods returned [ OK ]; Now performing reboot.

8

Update:

After reinstalling nvidia-gpu-firmware along with other nvidia and xorg-nvidia stuffs, and after force-rebuild akmods, dnf now detects version 595.71.05 as its update.
I’m now performing through ssh; Since the specific version had issues for me, I think I have to perform update on seat0.

It would be around 3~4PM TZ UTC+09:00. Will update the thread with results.

Target time updated due to personal situation.

9

Ran update through discover and akmods rebuild again; Now nvidia-smi recognizes the GPU and so does the kernel framebuffer. Thanks to the solution (simpler than speaking but still don’t know why it hasn’t do the transaction itself), and sorry if I bothered you all.

In short: it was NOT a defective driver. Rather, it didn’t akmod rebuild on transaction and made version mismatch which lead to defective state.