[BUG] AlmaLinux 10 - NVidia drivers installation not working on some systems

[DonnieAlex]

The documentation makes it easy to believe that 2 dnf commands would suffice to get the drivers up and running.
And most likely it is so, on a specific set of configurations.

However, personal workstations, tailored for game industry work as in my case, pose different challenges

Regular installation procedure as outlined in the docs:

sudo dnf install almalinux-release-nvidia-driver

sudo dnf install nvidia-open # <--- NOT THE MINIMAL, the minimal did not work for me.

sudo modprobe nvidia_drm OR reboot

at this point, my main desktop was screwed. Only one monitor worked, and at a VERY low resolution. During boot
it was complaining it failed to initialize nvidia_drm, as it couldn't find it.
my laptop instead did not show any visible issue. Running nvidia-smi showed no processes though.

activate persistence service or cards with multiple HDMI ports will revert to one port only after
sleep / logout / lock screen (like in my case)

sudo start nvidia-persistenced.service

sudo systemctl enable nvidia-persistenced.service

create or update this file -> /etc/dracut.conf.d/nvidia-gsp.conf by adding the following
the folder named after the driver version needs to be updated at every driver update!

install_items+=" /lib/firmware/nvidia/590.48.01/gsp_ga10x.bin " # notice the trailing spaces, they're required

rebuild initramfs!!

sudo dracut -f

reboot

at this point for my main workstation i did not need the following, nvidia-smi showed ALL graphic processes running on the GPU. On my laptop, instead, I did
The laptop was running on the integrated GPU all the time (nvidia-smi processes section showed no processes running) until apps would explicitly
request the nvidia GPU capabilities (benchmarking the Vulkan nvidia GPU showed the process in the tab created by nvidia-smi)
add the following environment variable to the file /etc/environment (assuming KDE installed)
devices names as they appear from the command lshw -short OR if the name is missing, the whole H/W path

KWIN_DRM_DEVICES=/dev/dri/{other_device_name}:/dev/dri/{nvidia_device_name}

My main Desktop info from fastfetch

---

OS: AlmaLinux 10.1 (Heliotrope Lion) x86_64
Host: Intel Z790 Workstation
Kernel: Linux 6.12.0-124.21.1.el10_1.x86_64
Display (Q32V4WG5): 2560x1440 @ 75 Hz in 32 [External]
Display (Q32V4WG5): 2560x1440 @ 75 Hz in 32 [External] *
DE: KDE Plasma 6.4.5
WM: KWin (Wayland)
CPU: Intel(R) Core(TM) i9-14900KS (32) @ 6.20 GHz
GPU: NVIDIA GeForce RTX 4090 [Discrete]
Memory: 3.37 GiB / 188.47 GiB (2%)
Swap: 0 B / 32.00 GiB (0%)
Disk (/): 4.39 GiB / 3.61 TiB (0%) - btrfs
Locale: en_GB.UTF-8

[DonnieAlex]

I've come across 2 more use cases.
On my laptop (an Optimus setup), I can remove the env var KWIN_DRM_DEVICES from the /etc/environment file and replace it with
DRI_PRIME=1
obtaining, as a result, that the integrated GPU runs all the time but as soon as a more intensive task comes along, it automatically passes it to the GPU, increasing battery life.

On my workstation, Maya 2025 and above (Qt6) has an issue with the viewport and the UVEditor (and possibly other editors) running an OpenGL based widget, rendering that portion of the screen completely black.
After extensive research and testing, I found out that adding the environment variable
__GLX_VENDOR_LIBRARY_NAME=mesa
to Maya (either in the .desktop entry, the Maya.env file or to the launch script) makes Maya load the Viewport2.0 correctly

Initialized VP2.0 renderer {
  Version : 2016.11.62.12. Feature Level 5.
  Adapter : zink Vulkan 1.4(NVIDIA GeForce RTX 4090 (NVIDIA_PROPRIETARY))
  .....
  Driver : 4.6 (Compatibility Profile) Mesa 25.0.7.
  API : OpenGL V.4.6.
  ....
}

as opposed to the default

Initialized VP2.0 renderer {
  Version : 2016.11.62.12. Feature Level 5.
  Adapter : NVIDIA GeForce RTX 4090/PCIe/SSE2
  ...
  Driver : 4.6.0 NVIDIA 590.48.01.
  API : OpenGL V.4.6.
  ....
}

and those widgets do NOT render black anymore, restoring software usability

[DonnieAlex]

bad news though... just noticed that openCL can't be initialized when using BOTH methods described in the previous comment (for Optimus setups on laptops and the env var on desktop)

OpenCL evaluator is attempting to initialize OpenCL.
Detected 1 OpenCL Platforms: 
 0: NVIDIA Corporation. NVIDIA CUDA. OpenCL 3.0 CUDA 13.1.112. ### looks promising, doesn't it?
 Supported extensions: cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_fp64 cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_icd cl_khr_gl_sharing cl_nv_compiler_options cl_nv_device_attribute_query cl_nv_pragma_unroll cl_nv_copy_opts cl_nv_create_buffer cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_device_uuid cl_khr_pci_bus_info cl_khr_external_semaphore cl_khr_external_memory cl_khr_external_semaphore_opaque_fd cl_khr_external_memory_opaque_fd cl_khr_semaphore 
OpenCL evaluator failed to find OpenCL device with vendor Advanced Micro Devices, Inc.. ### uh-oh... it's looking for an AMD card...

This means that two features in Maya won't work: XGen (hair creation system) and GPU Override (deformer nodes override piping them through GPU)

Although promising and an acceptable short-term fix to work with main Maya's features, this is still a patched up workaround and a fix needs to be implemented in the NVidia drivers.

[cachyosuser]

__GLX_VENDOR_LIBRARY_NAME=mesa fixed my viewport indeed. Sadly, when changing zoom level or dragging in graph editor, it won't update until the cursor moves to the viewport. Weirdly the graph updates when moving a keyframe.

[DonnieAlex]

hey cachyosuser! sorry to hear that this isn't quite working on your setup. If you're on CachyOS as your username implies, that might be the issue, with CachyOS having a game-performance oriented patched kernel. I'm on AlmaLinux and I'm not having the issues you're describing. All the previously black viewports (VP2.0, UVEditor and GraphEditor) work fine and render buttersmooth