IneffableParadise Hello, @IneffableParadise! 🐧😊
Thank you for performing all the tests and sharing your results!
I've carefully reviewed your OBS logs, and we now have enough information to focus on a much more specific issue.
The problem does not appear to be a missing NVIDIA driver. It is more likely related to NVIDIA's EGL/Wayland integration and the PipeWire screen-capture pipeline.
Your logs contain three particularly important messages:
Unable to open DRM render node.
Loading up OpenGL on adapter Mesa llvmpipe
[pipewire] buffer contains corrupted data
They also contain:
[obs-nvenc] Not running on NVIDIA GPU,
falling back to non-texture encoder
This tells us that OBS detects NVIDIA NVENC, but its OpenGL rendering is still falling back to the CPU. The screen-capture pipeline is also reporting corrupted buffers.
Since you have confirmed that X11 works correctly, I suggest the following approach.
1. Try running OBS through XWayland
This is the first solution I recommend testing.
XWayland allows X11 applications to run inside a Wayland session. This may bypass the problematic native EGL/Wayland rendering path.
First, close OBS completely.
Then run:
flatpak run --socket=x11 \
--env=QT_QPA_PLATFORM=xcb \
--env=__GLX_VENDOR_LIBRARY_NAME=nvidia \
com.obsproject.Studio
Open OBS, record a 30-second test video, and check whether the visual corruption is gone.
Next, go to Help → Log Files → View Current Log and look for:
Loading up OpenGL on adapter NVIDIA
If the log identifies NVIDIA instead of llvmpipe, hardware rendering has been restored.
This approach does not require switching your entire Plasma session to X11.
If it works, you can make the configuration permanent:
flatpak override --user \
--socket=x11 \
--env=QT_QPA_PLATFORM=xcb \
--env=__GLX_VENDOR_LIBRARY_NAME=nvidia \
com.obsproject.Studio
To remove the environment overrides later:
flatpak override --user \
--unset-env=QT_QPA_PLATFORM \
--unset-env=__GLX_VENDOR_LIBRARY_NAME \
com.obsproject.Studio
If you also wish to revoke the additional X11 socket permission:
flatpak override --user \
--nosocket=x11 \
com.obsproject.Studio
Only use the last command if you intentionally want to restrict X11 access.
2. Check the NVIDIA EGL libraries
I'd like to investigate one specific component: egl-wayland2.
This library provides a newer DMA-BUF-based EGL/Wayland implementation for NVIDIA GPUs and supports the NVIDIA 615 driver series.
Run:
pacman -Q nvidia-utils egl-wayland \
egl-wayland2 egl-gbm egl-x11 libglvnd
Check whether all the packages are installed.
If a package is missing, verify its availability in your configured BigLinux/Manjaro repositories.
For example:
pacman -Si egl-wayland2
If egl-wayland2 is missing and available from your official repositories, install it using:
sudo pacman -Syu egl-wayland2
Please avoid manually downloading packages from Arch Linux repositories or mixing NVIDIA driver versions.
After installing any missing libraries, restart your graphical session and test OBS again.
3. Test NVIDIA EGL directly under Wayland
Run:
eglinfo -B -p wayland
Then explicitly select NVIDIA's EGL implementation:
__EGL_VENDOR_LIBRARY_FILENAMES=/usr/share/glvnd/egl_vendor.d/10_nvidia.json eglinfo -B -p wayland
The 10_nvidia.json file must exist for the second test.
Ideally, the output should identify NVIDIA as the renderer.
If the first command reports llvmpipe, but the second reports NVIDIA, the issue may be related to automatic EGL vendor selection.
If the second command reports eglInitialize failed, we should investigate NVIDIA's Wayland EGL integration or its supporting libraries.
Regarding your earlier question: yes, there are two leading underscores in __GLX_VENDOR_LIBRARY_NAME.
You also need to set the variable on the same command line:
__GLX_VENDOR_LIBRARY_NAME=nvidia glxinfo -B
Running the assignment separately without exporting it does not pass that variable to the next command.
Also, glxinfo checks GLX/XWayland, while eglinfo lets us test the native EGL/Wayland path.
4. Check access to the DRM devices
OBS reported that it was unable to open a DRM render node.
First, check the host system:
ls -l /dev/dri/
id -nG
Then check what OBS can see inside its Flatpak sandbox:
flatpak run --command=sh \
com.obsproject.Studio \
-c 'ls -l /dev/dri/'
Inspect its permissions:
flatpak info --show-permissions \
com.obsproject.Studio
flatpak override --user --show \
com.obsproject.Studio
If the DRM devices are missing from the sandbox, try:
flatpak run --device=dri \
com.obsproject.Studio
Normally, OBS already has GPU device access, so there is no reason to change permissions unless the tests reveal an actual restriction.
Please do not use chmod 777 on GPU devices or disable Flatpak sandbox security.
5. Verify the results
After the changes:
- Start OBS and check that its renderer is no longer
llvmpipe.
- Create a new scene using Screen Capture (PipeWire).
- Record a 30–60 second test.
- Test NVIDIA NVENC H.264 again.
- Confirm that the visual corruption is gone and that PipeWire no longer reports corrupted buffers.
Please keep in mind that restoring hardware OpenGL rendering does not automatically prove that the PipeWire capture issue has been fixed. Both need to be verified.
Final thoughts
I also found a related KDE bug report where NVIDIA EGL/Wayland applications fall back to llvmpipe even though KWin itself is using the NVIDIA GPU.
There is also an OBS issue discussing NVIDIA/Wayland rendering problems and suggesting XWayland as a workaround.
References:
These reports suggest that the issue could involve NVIDIA's Wayland graphics integration, although we cannot yet confirm that your system is affected by exactly the same bug.
My recommendation is to start with the XWayland workaround. If successful, you may be able to continue using Plasma Wayland while running OBS with hardware acceleration.
If the issue persists, please share the results from steps 2, 3 and 4, along with a new OBS log.
In the meantime, Plasma X11 remains the confirmed working solution for your computer.
Thank you again for helping us investigate this issue and improve BigLinux! 🇧🇷🐧
Thank you for your time!
Ruscher