NVIDIA* Drivers

NVIDIA manufactures graphics processing units (GPU), also known as graphics cards.

NVIDIA devices on Linux* have two popular device driver options: the opensource drivers from the nouveau project or the proprietary drivers published by NVIDIA. The nouveau drivers are built into the Clear Linux* OS kernel and are loaded automatically at system boot if a compatible card is detected.

These instructions show how to use the proprietary NVIDIA drivers, which require a manual installation.

Warning

Software installed outside of swupd is not updated with Clear Linux OS updates and must be updated and maintained manually.

For example, the file /usr/lib/libGL.so conflicts with the file provided by the mesa package in Clear Linux OS and the file NVIDIA provides. If a Clear Linux OS update or repair overwrites these files, a reinstallation of the NVIDIA driver might be required.

Prerequisites

  • A Clear Linux OS system with a desktop installed

  • An NVIDIA device installed

Known issues

Systems with multiple graphics devices, including integrated graphics (iGPU), are known to be problematic.

Note

NVIDIA Optimus*

Some systems come with a hybrid graphics configuration for a balanced power and performance profile. This configuration is commonly found on laptops. NVIDIA Optimus technology, is designed to allow switching seamlessly between a NVIDIA device and another graphics devices sharing the same display.

Getting NVIDIA Optimus on Linux working well with both graphics devices adds an additional level of complexity with platform specific steps and may require additional software. Installation for systems with NVIDIA Optimus with both graphics devices operating is not covered by the scope of this documentation. As a simple workaround, some systems can disable one of the graphics devices or NVIDIA Optimus in the system firmware.

Note

The Long Term Support (LTS) kernel variant is more likely to be compatible with proprietary NVIDIA drivers.

See the Troubleshooting section for more known issues and solutions.

Installation

Configure workarounds

Some workarounds are required for the NVIDIA proprietary drivers to be usable and sustainable on Clear Linux OS.

  1. Remove the kernel command-line parameter intel_iommu=igfx_off or disable input–output memory management unit (IOMMU), also known as Intel® Virtualization Technology (Intel® VT) for Directed I/O (Intel® VT-d), in your system EFI/BIOS. See this GitHub report and the NVIDIA documentation on DMA issues for more information.

    The intel_iommu-igfx_off kernel parameter can be removed with the commands below:

    sudo mkdir -p /etc/kernel/cmdline-removal.d/
    echo "intel_iommu=igfx_off" | sudo tee /etc/kernel/cmdline-removal.d/intel-iommu.conf
    
  2. Create a custom systemd unit that overwrites the libGL library after every Clear Linux OS update with a pointer to the NVIDIA provided copy instead of the version provided by Clear Linux OS. These libraries conflict causing the NVIDIA driver to break when Clear Linux OS updates mesa. See the NVIDIA documentation on installed components for more information.

    1. Create a systemd service unit to overwrite the Clear Linux OS provided libGL.so.1 files with a symlink to the NVIDIA copies.

      sudo tee /etc/systemd/system/fix-nvidia-libGL-trigger.service > /dev/null <<'EOF'
      [Unit]
      Description=Fixes libGL symlinks for the NVIDIA proprietary driver
      BindsTo=update-triggers.target
      
      [Service]
      Type=oneshot
      ExecStart=/usr/bin/ln -sfv /opt/nvidia/lib/libGL.so.1 /usr/lib/libGL.so.1
      ExecStart=/usr/bin/ln -sfv /opt/nvidia/lib32/libGL.so.1 /usr/lib32/libGL.so.1
      EOF
      
    2. Reload the systemd manager configuration to pickup the new serivce.

      sudo systemctl daemon-reload
      
    3. Add the service as a depndency to the Clear Linux OS updates trigger causing the service to run after every update.

      sudo systemctl add-wants update-triggers.target fix-nvidia-libGL-trigger.service
      

Install DKMS

The Dynamic Kernel Module System (DKMS) allows the NVIDIA kernel modules to be automatically integrated when kernel updates occur in Clear Linux OS. Install the appropriate DKMS bundle using the instructions below:

Note

The Long Term Support (LTS) kernel variant is more likely to remain compatible between updates with NVIDIA drivers.

The kernel-native-dkms bundle provides the DKMS program and Linux kernel headers, which are placed under /usr/lib/modules/$(uname -r)/build/include/ and are required to compile kernel modules.

The kernel-native-dkms bundle also:

  • Adds a systemd update trigger (/usr/lib/systemd/system/dkms-new-kernel.service) to automatically run DKMS to rebuild modules after a kernel upgrade occurs with swupd update.

  • Disables kernel module signature verification by appending a kernel command-line parameter (module.sig_unenforce) from the /usr/share/kernel/cmdline.d/clr-ignore-mod-sig.conf file.

  • Adds a notification to the Message of the Day (MOTD) indicating kernel module signature verification is disabled.

Warning

We recommend that you always review the swupd update output to make sure kernel modules were successfully rebuilt against the new kernel. This is especially important for systems where a successful boot relies on a kernel module.

Install the kernel-native-dkms or kernel-lts-dkms bundle:

  1. Determine which kernel variant is running on Clear Linux OS. Only the native and lts kernels are enabled to build and load out-of-tree kernel modules with DKMS.

    $ uname -r
    5.XX.YY-ZZZZ.native
    

    Ensure .native or .lts is in the kernel name.

  2. Install the DKMS bundle corresponding to the installed kernel. Use kernel-native-dkms for the native kernel or kernel-lts-dkms for the lts kernel.

    sudo swupd bundle-add kernel-native-dkms
    

    or

    sudo swupd bundle-add kernel-lts-dkms
    
  3. Update the Clear Linux OS bootloader and reboot.

    sudo clr-boot-manager update
    reboot
    

Download the NVIDIA drivers

  1. Identify the NVIDIA GPU model that is installed.

    sudo lshw -C display
    
  2. Go to the NVIDIA Driver Downloads website . Search for and download the appropriate driver based on the NVIDIA GPU model you have with Linux 64-bit selected as the Operating System.

    wget https://download.nvidia.com/XFree86/Linux-x86_64/<VERSION>/NVIDIA-Linux-x86_64-<VERSION>.run
    

    If you already know the appropriate driver version for your device, you can also obtain a download link directly from one of the links below:

Disable the nouveau driver

The proprietary NVIDIA driver is incompatible with the nouveau driver and must be disabled before installation can continue.

  1. Disable the nouveau driver by creating a blacklist file under /etc/modprobe.d and reboot.

    sudo mkdir /etc/modprobe.d
    
    printf "blacklist nouveau \noptions nouveau modeset=0 \n" | sudo tee --append /etc/modprobe.d/disable-nouveau.conf
    
  2. Reboot the system and log back in. It is normal for the graphical environment not to start without the NVIDIA driver loaded.

Configure alternative software paths

The NVIDIA installer is directed to install files under /opt/nvidia as much as possible to keep its contents isolated from the rest of the Clear Linux OS system files under /usr. The dynamic linker and X server must be configured to use the content under /opt/nvidia.

  1. Configure the dynamic linker to look for and to cache shared libraries under /opt/nvidia/lib and /opt/nvidia/lib32 in addition to the default paths.

    echo "include /etc/ld.so.conf.d/*.conf" |  sudo tee --append /etc/ld.so.conf
    
    sudo mkdir /etc/ld.so.conf.d
    printf "/opt/nvidia/lib \n/opt/nvidia/lib32 \n" | sudo tee --append /etc/ld.so.conf.d/nvidia.conf
    
  2. Reload the dynamic linker run-time bindings and library cache.

    sudo ldconfig
    
  3. Create a Xorg configuration file to search for modules under /opt/nvidia in addition to the default path.

    sudo mkdir -p /etc/X11/xorg.conf.d/
    
    sudo tee /etc/X11/xorg.conf.d/nvidia-files-opt.conf > /dev/null <<'EOF'
    Section "Files"
            ModulePath      "/usr/lib64/xorg/modules"
            ModulePath      "/opt/nvidia/lib64/xorg/modules"
    EndSection
    EOF
    

Install the NVIDIA drivers

  1. A terminal not running on /dev/tty1 is useful to view uninterrupted installation progress. Switch to a secondary virtual terminal by pushing CTRL + ALT + F2 or remotely login over SSH.

  2. Navigate to the directory where the NVIDIA installer was downloaded. In this example, it was saved in the Downloads folder.

    cd ~/Downloads/
    
  3. Run the installer with the advanced options below.

    sudo sh NVIDIA-Linux-x86_64-<VERSION>.run \
    --utility-prefix=/opt/nvidia \
    --opengl-prefix=/opt/nvidia \
    --compat32-prefix=/opt/nvidia \
    --compat32-libdir=lib32 \
    --x-prefix=/opt/nvidia \
    --x-module-path=/opt/nvidia/lib64/xorg/modules \
    --x-library-path=/opt/nvidia/lib64 \
    --x-sysconfig-path=/etc/X11/xorg.conf.d \
    --documentation-prefix=/opt/nvidia \
    --application-profile-path=/etc/nvidia/nvidia-application-profiles-rc.d \
    --no-precompiled-interface \
    --no-nvidia-modprobe \
    --no-distro-scripts \
    --force-libglx-indirect \
    --glvnd-egl-config-path=/etc/glvnd/egl_vendor.d \
    --egl-external-platform-config-path=/etc/egl/egl_external_platform.d  \
    --dkms \
    --silent
    
  4. The graphical interface may automatically start after the NVIDIA driver is loaded. Return to the working terminal and log back in if necessary.

  5. Confirm that the NVIDIA kernel modules are loaded.

    lsmod | grep ^nvidia
    
  6. Optional: Create a link for the nvidia-settings desktop entry to ~/.local/share/applications so that it appears in the launcher for easy access.

    ln -sv /opt/nvidia/share/applications/nvidia-settings.desktop $HOME/.local/share/applications
    

Updating

The proprietary NVIDIA drivers are installed manually outside of swupd and must be updated manually when needed.

Updating the NVIDIA drivers follows the same steps as initial installation, however the desktop environment must first be stopped so that the drivers are not in use.

  1. Follow the steps in the Download the NVIDIA drivers section to get the latest NVIDIA drivers.

  2. Temporarily set the default boot target to the multi-user, which is a non-graphical runtime.

    sudo systemctl set-default multi-user.target
    
  3. Reboot the system and log back in. It is normal for the graphical environment not to start.

  4. Follow the steps in the Install the NVIDIA Drivers section to update the NVIDIA drivers. This installation will overwrite the previous NVIDIA drivers and files.

  5. Set the default boot target back to the graphical target.

    sudo systemctl set-default graphical.target
    
  6. Reboot the system and log back in.

  7. Trigger a flatpak update to download the runtime corresponding with the new NVIDIA drivers for the flatpak apps that require it.

    Important

    Some flatpak applications won’t start after updating the NVIDIA drivers until the flatpak runtime is updated with the corresponding driver version.

    flatpak update
    

Uninstallation

The NVIDIA drivers and associated software can be uninstalled and nouveau driver restored with the instructions in this section.

  1. Remove the files created for workarounds.

    sudo rm /etc/kernel/cmdline-removal.d/intel-iommu.conf
    sudo rm /etc/systemd/system/fix-nvidia-libGL-trigger.service
    sudo rm /etc/systemd/system/update-triggers.target.wants/fix-nvidia-libGL-trigger.service
    sudo systemctl daemon-reload
    
  2. Remove the modprobe.d file that prevents nouveau from loading.

    sudo rm /etc/modprobe.d/disable-nouveau.conf
    
  3. Remove the nvidia.conf file so that dynamic linker does not look for cached libraries under /opt/nvidia/lib and /opt/nvidia/lib32.

    sudo rm /etc/ld.so.conf.d/nvidia.conf
    sudo ldconfig
    

    Optionally, restore ld.so.conf to default if no other configuration files under /etc/ld.so.conf.d needs to be included.

    sudo sed -i '/^include \/etc\/ld\.so\.conf\.d\/\*\.conf$/d' /etc/ld.so.conf
    
  4. Remove the xorg.conf.d file that adds a search path for X modules.

    sudo rm /etc/X11/xorg.conf.d/nvidia-files-opt.conf
    
  5. Remove the nvidia-settings desktop entry file if it was linked to ~/.local/share/applications.

    unlink -v $HOME/.local/share/applications/nvidia-settings.desktop
    
  6. Run the nvidia-uninstall command.

    sudo /opt/nvidia/bin/nvidia-uninstall
    
  7. Follow the prompts on the screen and reboot the system.

Troubleshooting

  • The NVIDIA driver places installer and uninstaller logs under /var/log/nvidia-install and /var/log/nvidia-uninstall.

  • NVIDIA-Linux-x86_64-<VERSION>.run --advanced-options shows many parameters to control installation behavior.

  • NVIDIA-Linux-x86_64-<VERSION>.run --extract-only extracts installation files into a directory named NVIDIA-Linux-x86_64-<VERSION>.

  • The X server logs under /var/log/X* contain useful information about display and driver loading. Check all the files and timestamps when troubleshooting.

  • The DKMS build logs under /var/lib/dkms/nvidia* contain information about kernel module builds which can be useful if the NVIDIA driver breaks between kernel upgrades.

No display or blank screen

Check to see if the display has come up on another graphics device, including the integrated graphics device.

You might get a black screen or the login screen might not come up after installing the NVIDIA drivers until an Xorg configuration has been defined for your monitors.

“Oh no! Something has gone wrong” GNOME* crash

NVIDIA driver GNOME crash on Clear Linux OS

NVIDIA driver GNOME crash dialogue on Clear Linux OS.

There have been reports of GNOME crashing with an “Oh no! Something has gone wrong” error message with NVIDIA drivers installed while other graphics devices are enabled.

Try disabling other graphics devices, including integrated graphics, in your system’s EFI/BIOS.

Slow boot times

There have been reports of slow boot times with NVIDIA drivers installed. Normally, when GDM detects NVIDIA proprietary drivers, it will disable Wayland and enable X11. Should GDM fail to disbale Wayland, it may results in slow boot times, according to this GitHub reprot.

To manually disable Wayland:

sudo tee /etc/gdm/custom.conf > /dev/null <<'EOF'
[daemon]
WaylandEnable=false
EOF

Brightness control

If you can’t control the screen brightness with the NVIDIA driver installed, try one of the solutions below:

  • Add a kernel parameter acpi_osi= which disables the ACPI Operating System Identification function. Some system firmware may manipulate brightness control keys based on the reported operating system. Disabling the identification mechanism can cause the system firmware to expose brightness controls that are recognizable in Linux.

    sudo mkdir -p /etc/kernel/cmdline.d
    echo "acpi_osi=" | sudo tee /etc/kernel/cmdline.d/acpi-backlight.conf
    sudo clr-boot-manager update
    
  • Add a kernel parameter for the nvidia driver: NVreg_EnableBacklightHandler=1. This handler overrides the ACPI-based one provided by the video.ko kernel module. This option is available with NVIDIA driver version 387.22 and above.

    sudo mkdir -p /etc/kernel/cmdline.d
    echo "nvidia.NVreg_EnableBacklightHandler=1" | sudo tee /etc/kernel/cmdline.d/nvidia-backlight.conf
    sudo clr-boot-manager update
    
  • Add the EnableBrightnessControl=1 options to the Device section of your xorg config. Below is an example:

    sudo mkdir -p /etc/X11/xorg.conf.d/
    
    sudo tee /etc/X11/xorg.conf.d/nvidia-brightness.conf > /dev/null <<'EOF'
    Section "Device"
        Identifier     "Device0"
        Driver         "nvidia"
        Option         "RegistryDwords" "EnableBrightnessControl=1"
    EndSection
    EOF