nvidia-driver-on-clear-linux

A how-to NVIDIA proprietary driver installation on Clear Linux OS.

Preparation

Starting fresh? Obtain the current image from the Clear Linux Downloads page.

Clear Linux OS installation and updates

Depending on the CL release, the open-source nouveau driver may not work with recent NVIDIA graphics (3000 series or newer). The solution is to install the OS in text mode. Press the letter e on the boot screen and prepend nomodeset 3 with a space to the list of kernel arguments. The OS will boot into multi-user mode and prevent the nouveau driver from loading. Press enter to boot the OS. Instructions are provided on the screen for running the installer.

During setup, remember to enter [A] Advanced options. Select "Kernel Command Line" and enter nomodeset 3 to "Add Extra Arguments". Optionally, go back to the prior screen and choose the lts (recommended) or native kernel under "Kernel Selection". Also disable automatic OS updates if desired.

NVIDIA driver installation

git clone https://github.com/marioroy/nvidia-driver-on-clear-linux
cd nvidia-driver-on-clear-linux

First, run the pre-installer script. The update argument is needed on prior Clear Linux installations or to refresh the configuration files under /etc/. Run the pre-installer script subsequently, without an argument, to switch the target from graphical to multi-user (text-mode).

$ bash ./pre-install-driver help
Usage: pre-install-driver [ update ]

$ git pull
$ bash ./pre-install-driver update
$ bash ./pre-install-driver

Next, run the driver installer script. Specifying latest installs the latest production release. Check first before installation. Or specify the desired version or path to the installer file.

$ bash ./install-driver help
Usage: install-driver latest|550|560|565|vulkan|<pathname>

$ bash ./install-driver 550
$ sudo reboot

NVIDIA CUDA Toolkit installation

Installing the CUDA Toolkit is optional. The "auto" argument is preferred and will install the version suitable for the display driver. If the display driver is not in the table, then the script will fetch the latest CUDA run-file. If unsure, install the CUDA version matching the display driver or a lower version supported by the application (e.g. Blender).

$ bash ./install-cuda help
Usage: install-cuda auto|latest|12.6|12.4|12.2|11.8|<pathname>

$ bash ./install-cuda auto    # or path to run file
$ bash ./install-cuda ~/Downloads/cuda_12.4.1_550.54.15_linux.run

Update ~/.bashrc so that it can find the nvcc command.

export CUDA_HOME=/opt/cuda
export PATH=$CUDA_HOME/bin:$PATH
export LD_LIBRARY_PATH=$CUDA_HOME/lib64:$LD_LIBRARY_PATH

Or if you prefer, use module to prepend the CUDA path. First time? This may require logging out and back in, or open a new terminal window.

$ module load cuda
$ nvcc --version
$ module unload cuda

Q) Why specify the auto argument to install-cuda or lower version than the display driver?

A) Using a later CUDA Toolkit release than the display driver may cause some CUDA programs to emit an error, "the provided PTX was compiled with an unsupported toolchain".

Q) What are my options if the install-cuda script is no longer current?

A) Notifying the author is one option. The other option is to visit the CUDA archive URL, provided at the bottom of the page. Determine the display driver version from the filename. Then visit the driver archive and scroll to the bottom of the page. Click on the latest matching the base version. Install the driver and CUDA Toolkit by providing the path to install-driver and install-cuda respectively.

Enable hardware acceleration

Hardware video acceleration requires the NVDEC back-end VA-API driver. See the HWAccel folder. It provides instructions for building the VA driver and configuration file for Firefox.

Updating the NVIDIA driver

First, ensure you have the latest by running git pull followed by pre-install-driver with the update argument. You may stop here if all you want to do is refresh the NVIDIA-related configuration files from any upstream updates.

$ git pull
$ bash ./pre-install-driver update

The output will be much smaller for pre-install-driver without an argument since it will skip completed sections. Run the pre-installer script regardless to switch the boot target to text mode.

Run install-driver latest or acquire the run-file from NVIDIA and save it locally.

$ bash ./pre-install-driver

$ bash ./install-driver latest    # or path to run file
$ bash ./install-driver ~/Downloads/NVIDIA-Linux-x86_64-550.135.run
$ sudo reboot

Update the CUDA Toolkit, if installed, to install the version suitable for the display driver and supported by the application.

[ -d /opt/cuda ] && bash ./install-cuda auto

Uninstallation

The NVIDIA software can be uninstalled and nouveau driver restored. Reinstall the NVIDIA driver if the nouveau driver is failing for your graphics (e.g. NVIDIA 3000+ series).

$ bash ./uninstall-cuda
$ bash ./uninstall-driver
$ sudo reboot

Troubleshooting

Installing a kernel manually? The check-kernel-dkms script installs missing dkms bundles and runs dkms autoinstall per each kernel on the system.

bash ./check-kernel-dkms

Running Wayland and x11 apps are crashing? See issue. Try setting a system-wide variable and reboot.

__EGL_VENDOR_LIBRARY_FILENAMES=/usr/share/glvnd/egl_vendor.d/10_nvidia.json

Running a NVIDIA Optimus laptop? See CONFIGURATION STEPS or search for dbus in /usr/share/doc/NVIDIA_GLX-1.0/README.txt. Undo the steps if nvidia-powerd reports no matching GPU found. Refer to: Failure with the nvidia-powerd service and Video decode does not work after exiting sleep.

See also: Integrated GPU is not (fully) used / Responsiveness issue.

Experiencing stutter or tearing when moving windows? Lanuch NVIDIA Settings and enable "Force Full Composition Pipeline". See also, Difference between Force Full Composition Pipeline and Force Composition Pipeline.

Note: Skip this section on Optimus-based laptops. Instead, see the prior URL.

1. sudo chown $USER /etc/X11  # this is done by the install-driver script
2. Go to NVIDIA Settings > X Server Display Configuration > Advanced...
3. Enable "Force Full Composition Pipeline" per each monitor
4. Click "Apply"
5. Click "Save to X Configuration File" to persist the change

Add the UseNvKmsCompositionPipeline option to the Screen section in /etc/X11/xorg.conf; to minimize power-consumption on idle when enabling CompositionPipeline.

Option "UseNvKmsCompositionPipeline" "Off"

Using a HiDPI display? In GNOME, run gnome-tweaks and adjust "Scaling Factor" on the Fonts pane. For Wayland, restore text-scaling-factor back to 1.0 and enable an experimental feature scale-monitor-framebuffer. Log out for the settings to take effect. Go to GNOME settings > Displays and set the scale accordingly, for the display to 100%, 125%, 150%, ... 400%.

gsettings set org.gnome.desktop.interface text-scaling-factor 1.0
gsettings set org.gnome.mutter experimental-features "['scale-monitor-framebuffer']"

Installing KDE? Important, remove the desktop-autostart bundle to not autostart GDM resulting in black screen.

sudo swupd bundle-remove desktop-autostart
sudo swupd bundle-add desktop-kde
sudo swupd bundle-add desktop-kde-apps   # optional

See also

• Announcement; tips and solutions

• Download Driver Archive, Unix Driver Archive, Vulkan Driver Support

• CUDA Toolkit Archive, Documentation, Samples

• CUDA Redistributable Driver Archive