Translation(s): English - Español - Français - Русский - 简体中文


NVIDIA Proprietary Driver

Nvidia Graphics Drivers are partly secret source (closed source/proprietary) software, owned by a for profit corporation and not supported by Debian. For those interested in stronger security and stronger privacy, it is suggested to consider using an alternative to Nvidia Graphics Drivers. Which is both fully libre source (open source) and supported by Debian. Such as Mesa.

This page describes how to install the NVIDIA proprietary display driver on Debian systems.

Commands in this article prefixed with a # indicate they must be run as root. Replace this character with sudo or switch user to root in your terminal beforehand as necessary.

NOTE: For Apple systems, follow these steps first to prevent a black screen after installing the drivers: https://askubuntu.com/a/613573/134848

Identification

The NVIDIA graphics processing unit (GPU) series/codename of an installed video card can usually be identified using the lspci command. For example:

See HowToIdentifyADevice/PCI for more information. The PCI ID can be used to verify device support.

Note: if this lspci command returns more than one line of output, one of two things are possible:

nvidia-detect

The nvidia-detect script (found in the nvidia-detect package in the non-free section) can also be used to identify the GPU and the recommended driver package to install:

Note: if the recommended driver is for example only "nvidia-tesla-470-driver", then adapt all following instructions to nvidia-tesla-470* packages, and don't use the corresponding nvidia* or nvidia-tesla* packages, as this will likely result in conflicts and driver mismatches.

Desktop Drivers

The proprietary "NVIDIA Accelerated Linux Graphics Driver" provides optimized hardware acceleration of OpenGL and Vulkan applications through either Xorg or Wayland. It is a binary-only driver requiring a Linux kernel module for its use.

Multiple precompiled driver versions are available for Debian Unstable "Sid":

Only one precompiled driver version is available for Debian 12 "Bookworm":

Multiple precompiled driver versions are available for Debian 11 "Bullseye":

Multiple precompiled driver versions are available for Debian 10 "Buster":

Multiple precompiled driver versions are available for Debian 9 "Stretch":

All driver versions up to, and including, the 418-series, are only available for the x86, x86-64, and 32-bit ARMv7 architectures (Debian i386, AMD64, and ARMHF ports respectively).

The 450-series and newer has dropped support for 32-bit architectures, now only supporting x86-64 and ARMv8 (Debian AMD64 and ARM64 ports respectively).


Prerequisites

Kernel headers

Before installing the drivers, you must obtain the proper kernel headers for the NVIDIA driver to build with.

For a typical 64-bit system using the default kernel, you can simply run:

# apt install linux-headers-amd64

For 32-bit systems with the non-PAE kernel, you'd instead install:

# apt install linux-headers-686

Or, for 32-bit systems with the PAE kernel:

# apt install linux-headers-686-pae

If you're using the kernel from Debian Backports, you must run the same command but with the -t flag followed by the name of your backports source. For instance, if you're using backports on a 64-bit Debian 10 system, you might run:

# apt install -t buster-backports linux-headers-amd64


Kernel

In some cases, if you're aiming to install the bleeding-edge version of the NVIDIA driver from Debian Backports, you may also need to install the kernel from backports to match it. For Debian 10, you might do this with:

# apt install -t buster-backports linux-image-amd64

Exchange "buster-backports" with your own version's backports repository as necessary.


Installation

Debian Unstable "Sid"

Version 535.183.01

For support of GeForce 700 series and newer GPUs (supported devices). For older devices, see Version 390 (legacy GPUs).

  1. Add "contrib", "non-free" and "non-free-firmware" components to /etc/apt/sources.list, for example:

    # Debian Sid
    deb http://deb.debian.org/debian/ sid main contrib non-free non-free-firmware
  2. If you have SecureBoot enabled: you need to enroll your machine owner's key (MOK) to use DKMS modules. Detailed instructions are available here. It's recommended to do this before installing nvidia-driver so that you do not have to rebuild the kernel image.

  3. If your system uses dracut, you must tell dracut to include modprobe configuration files:

    • Make a dracut configuration file /etc/dracut.conf.d/10-nvidia.conf (you can actually name it anything you like, as long as it ends in .conf) with the following contents:

      install_items+=" /etc/modprobe.d/nvidia-blacklists-nouveau.conf /etc/modprobe.d/nvidia.conf /etc/modprobe.d/nvidia-options.conf "
    • Note the spaces between quotes and characters.
    • The modprobe.d files referenced here will be added by the nvidia-driver package.

  4. Update the list of available packages, then we can install the nvidia-driver package, plus the necessary firmware:

    # apt update
    # apt install nvidia-driver firmware-misc-nonfree

    DKMS will build the nvidia module for your system, via the nvidia-kernel-dkms package.

  5. Restart your system to load the new driver.

Version 390.144

For support of GeForce 400 series and newer GPUs (supported devices).

  1. Add "contrib", "non-free" and "non-free-firmware" components to /etc/apt/sources.list, for example:

    # Debian Sid
    deb http://deb.debian.org/debian/ sid main contrib non-free non-free-firmware
  2. If you have SecureBoot enabled: you need to enroll your machine owner's key (MOK) to use DKMS modules. Detailed instructions are available here. It's recommended to do this before installing nvidia-driver so that you do not have to rebuild the kernel image.

  3. If your system uses dracut, you must tell dracut to include modprobe configuration files:

    • Make a dracut configuration file /etc/dracut.conf.d/10-nvidia.conf (you can actually name it anything you like, as long as it ends in .conf) with the following contents:

      install_items+=" /etc/modprobe.d/nvidia-blacklists-nouveau.conf /etc/modprobe.d/nvidia.conf /etc/modprobe.d/nvidia-options.conf "
    • Note the spaces between quotes and characters.
    • The modprobe.d files referenced here will be added by the nvidia-driver package.

  4. Update the list of available packages, then we can install the nvidia-legacy-390xx-driver package, plus the necessary firmware:

    # apt update
    # apt install nvidia-legacy-390xx-driver firmware-misc-nonfree

    DKMS will build the nvidia module for your system, via the nvidia-legacy-390xx-kernel-dkms package.

  5. Restart your system to load the new driver.

Version 340.108

For support of GeForce 8 series through GeForce 300 series GPUs(supported devices).

Use of the 340-series driver is strongly discouraged. It is not included in stable releases of Debian anymore, has serious unfixable security vulnerabilities, and may not be updated for new kernels in a timely manner. You are highly recommended to use the built-in Nouveau driver if security is a priority.

  1. Add "contrib", "non-free" and "non-free-firmware" components to /etc/apt/sources.list, for example:

    # Debian Sid
    deb http://deb.debian.org/debian/ sid main contrib non-free non-free-firmware
  2. If you have SecureBoot enabled: you need to enroll your machine owner's key (MOK) to use DKMS modules. Detailed instructions are available here. It's recommended to do this before installing nvidia-driver so that you do not have to rebuild the kernel image.

  3. If your system uses dracut, you must tell dracut to include modprobe configuration files:

    • Make a dracut configuration file /etc/dracut.conf.d/10-nvidia.conf (you can actually name it anything you like, as long as it ends in .conf) with the following contents:

      install_items+=" /etc/modprobe.d/nvidia-blacklists-nouveau.conf /etc/modprobe.d/nvidia.conf /etc/modprobe.d/nvidia-options.conf "
    • Note the spaces between quotes and characters.
    • The modprobe.d files referenced here will be added by the nvidia-driver package.

  4. Update the list of available packages, then we can install the nvidia-legacy-340xx-driver package, plus the necessary firmware:

    # apt update
    # apt install nvidia-legacy-340xx-driver firmware-misc-nonfree

    DKMS will build the nvidia module for your system, via the nvidia-legacy-340xx-kernel-dkms package.

  5. Restart your system to load the new driver.


Debian 12 "Bookworm"

Version 535.183.01

For support of GeForce 700 series and newer GPUs (supported devices). For older devices, you must use nouveau, which should be already installed and in use.

  1. Add "contrib", "non-free" and "non-free-firmware" components to /etc/apt/sources.list, for example:

    # Debian Bookworm
    deb http://deb.debian.org/debian/ bookworm main contrib non-free non-free-firmware
  2. If you have SecureBoot enabled: you need to enroll your machine owner's key (MOK) to use DKMS modules. Detailed instructions are available here. It's recommended to do this before installing nvidia-driver so that you do not have to rebuild the kernel image.

  3. If your system uses dracut, you must tell dracut to include modprobe configuration files:

    • Make a dracut configuration file /etc/dracut.conf.d/10-nvidia.conf (you can actually name it anything you like, as long as it ends in .conf) with the following contents:

      install_items+=" /etc/modprobe.d/nvidia-blacklists-nouveau.conf /etc/modprobe.d/nvidia.conf /etc/modprobe.d/nvidia-options.conf "
    • Note the spaces between quotes and characters.
    • The modprobe.d files referenced here will be added by the nvidia-driver package.

  4. Update the list of available packages, then we can install the nvidia-driver package, plus the necessary firmware:

    # apt update
    # apt install nvidia-driver firmware-misc-nonfree

    DKMS will build the nvidia module for your system, via the nvidia-kernel-dkms package.

  5. Restart your system to load the new driver.
  6. If you plan to use wayland desktop (which is a default in Debian 12 "Bookworm") you may need to check and correct the following:

    • Check if the following command returns Y:

      # cat /sys/module/nvidia_drm/parameters/modeset
    • If this command returns N you need to set nvidia-drm modeset by adding options nvidia-drm modeset=1 to /etc/modprobe.d/nvidia-options.conf. For example:

      # echo "options nvidia-drm modeset=1" >> /etc/modprobe.d/nvidia-options.conf
    • Then reboot your system (check again if modeset is properly set, as described above) and Nvidia card should be properly recognized and used under wayland desktop.


Debian 11 "Bullseye"

Version 470.129.06

For support of GeForce 600 series and newer GPUs (supported devices). For older devices, see Version 390 (legacy GPUs).

  1. Add "contrib" and "non-free" components to /etc/apt/sources.list, for example:

    # Debian 11 "Bullseye"
    deb http://deb.debian.org/debian/ bullseye main contrib non-free
  2. If you have SecureBoot enabled: you need to enroll your machine owner's key (MOK) to use DKMS modules. Detailed instructions are available here. It's recommended to do this before installing nvidia-driver so that you do not have to rebuild the kernel image.

  3. Update the list of available packages, then we can install the nvidia-driver package, plus the necessary firmware:

    # apt update
    # apt install nvidia-driver firmware-misc-nonfree

    DKMS will build the nvidia module for your system, via the nvidia-kernel-dkms package.

  4. Restart your system to load the new driver.

Version 390.144

For support of GeForce 400 series and newer GPUs (supported devices).

  1. Add "contrib" and "non-free" components to /etc/apt/sources.list, for example:

    # Debian 11 "Bullseye"
    deb http://deb.debian.org/debian/ bullseye main contrib non-free
  2. If you have SecureBoot enabled: you need to enroll your machine owner's key (MOK) to use DKMS modules. Detailed instructions are available here. It's recommended to do this before installing nvidia-driver so that you do not have to rebuild the kernel image.

  3. Update the list of available packages, then we can install the nvidia-legacy-390xx-driver package, plus the necessary firmware:

    # apt update
    # apt install nvidia-legacy-390xx-driver firmware-misc-nonfree

    DKMS will build the nvidia module for your system, via the nvidia-legacy-390xx-kernel-dkms package.

  4. Restart your system to load the new driver.


Debian 10 "Buster"

Version 460.73.01 (via buster-backports)

For support of GeForce 600 series and newer GPUs (supported devices). For older devices, see Version 390 (legacy GPUs) and Version 340 (legacy GPUs).

  1. Add buster-backports as an additional new line to your /etc/apt/sources.list, for example:

    # buster-backports
    deb http://deb.debian.org/debian buster-backports main contrib non-free
  2. If you have SecureBoot enabled: you need to enroll your machine owner's key (MOK) to use DKMS modules. Detailed instructions are available here. It's recommended to do this before installing nvidia-driver so that you do not have to rebuild the kernel image.

  3. Update the list of available packages, then we can install the nvidia-driver package, plus the necessary firmware, from the backports repository:

    # apt update
    # apt install -t buster-backports nvidia-driver firmware-misc-nonfree

    DKMS will build the nvidia module for your system, via the nvidia-kernel-dkms package.

  4. Reboot your system to load the updated driver.

Version 418.197.02

For support of GeForce 600 series and newer GPUs (supported devices). For older devices, see Version 390 (legacy GPUs) and Version 340 (legacy GPUs).

  1. Add "contrib" and "non-free" components to /etc/apt/sources.list, for example:

    # Debian 10 "Buster"
    deb http://deb.debian.org/debian/ buster main contrib non-free
  2. If you have SecureBoot enabled: you need to enroll your machine owner's key (MOK) to use DKMS modules. Detailed instructions are available here. It's recommended to do this before installing nvidia-driver so that you do not have to rebuild the kernel image.

  3. Update the list of available packages, then we can install the nvidia-driver package, plus the necessary firmware:

    # apt update
    # apt install nvidia-driver firmware-misc-nonfree

    DKMS will build the nvidia module for your system, via the nvidia-kernel-dkms package.

  4. Restart your system to load the new driver.

Version 390.138 (legacy GPUs)

For support of GeForce 400 series and newer GPUs (supported devices).

  1. Add "contrib" and "non-free" components to /etc/apt/sources.list, for example:

    # Debian 10 "Buster"
    deb http://deb.debian.org/debian/ buster main contrib non-free
  2. If you have SecureBoot enabled: you need to enroll your machine owner's key (MOK) to use DKMS modules. Detailed instructions are available here. It's recommended to do this before installing nvidia-driver so that you do not have to rebuild the kernel image.

  3. Update the list of available packages, then we can install the nvidia-legacy-390xx-driver package, plus the necessary firmware:

    # apt update
    # apt install nvidia-legacy-390xx-driver firmware-misc-nonfree

    DKMS will build the nvidia module for your system, via the nvidia-legacy-390xx-kernel-dkms package.

  4. Restart your system to load the new driver.

Version 340.108 (legacy GPUs)

For support of GeForce 8 series through GeForce 300 series GPUs. (supported devices).

  1. Add "contrib" and "non-free" components to /etc/apt/sources.list, for example:

    # Debian 10 "Buster"
    deb http://deb.debian.org/debian/ buster main contrib non-free
  2. If you have SecureBoot enabled: you need to enroll your machine owner's key (MOK) to use DKMS modules. Detailed instructions are available here. It's recommended to do this before installing nvidia-driver so that you do not have to rebuild the kernel image.

  3. Update the list of available packages, then we can install the nvidia-legacy-340xx-driver package:

    # apt update
    # apt install nvidia-legacy-340xx-driver

    DKMS will build the nvidia module for your system, via the nvidia-legacy-340xx-kernel-dkms package.

  4. After, create an Xorg server configuration file and then restart your system to enable the nouveau blacklist.


Debian 9 "Stretch"

As of stretch, you don't need nvidia-xconfig anymore, and a xorg.conf file is not needed either in most situations. Also, the 340 series has been forked into its own series of packages to support older cards.

In some situations running nvidia-xconfig is still required for screen-locking and suspend/resume to work properly (922679 Xfce/lightdm/light-locker)

Version 418.152 (via stretch-backports)

For support of GeForce 700 series and newer GPUs (supported devices). For older devices, see Version 340 (legacy GPUs) and Version 304 (legacy GPUs).

  1. Add stretch-backports as an additional new line to your /etc/apt/sources.list, for example:

    # stretch-backports
    deb http://deb.debian.org/debian stretch-backports main contrib non-free
  2. Update the list of available packages, then we can install the nvidia-driver package, plus the necessary firmware, from the backports repository:

    # apt update
    # apt install -t stretch-backports nvidia-driver firmware-misc-nonfree

    DKMS will build the nvidia module for your system, via the nvidia-kernel-dkms package.

  3. Restart your system to enable the nouveau blacklist.

Version 390.138

For support of GeForce 400 series and higher GPUs (supported devices). For older devices, see Version 340 (legacy GPUs) and Version 304 (legacy GPUs).

  1. Add "contrib" and "non-free" components to /etc/apt/sources.list, for example:

    # Debian 9 "Stretch"
    deb http://deb.debian.org/debian/ stretch main contrib non-free
  2. Update the list of available packages, then we can install the nvidia-driver package, plus the necessary firmware:

    # apt update
    # apt install nvidia-driver firmware-misc-nonfree

    DKMS will build the nvidia module for your system, via the nvidia-kernel-dkms package.

  3. Restart your system to enable the nouveau blacklist.

Version 340.108 (legacy GPUs)

For support of GeForce 8 series through GeForce 300 series GPUs (supported devices).

  1. Add "contrib" and "non-free" components to /etc/apt/sources.list, for example:

    # Debian 9 "Stretch"
    deb http://deb.debian.org/debian/ stretch main contrib non-free
  2. Update the list of available packages, then we can install the nvidia-legacy-340xx-driver package:

    # apt update
    # apt install nvidia-legacy-340xx-driver

    DKMS will build the nvidia module for your system, via the nvidia-legacy-340xx-kernel-dkms package.

    After, create an Xorg server configuration file and then restart your system to enable the nouveau blacklist.

Version 304.137 (legacy GPUs)

For support of GeForce 6 series and GeForce 7 series GPUs (supported devices).

  1. Add "contrib" and "non-free" components to /etc/apt/sources.list, for example:

    # Debian 9 "Stretch"
    deb http://deb.debian.org/debian/ stretch main contrib non-free
  2. Update the list of available packages, then we can install the nvidia-legacy-304xx-driver package:

# apt install nvidia-legacy-304xx-driver


Installing 32-bit libraries on a 64-bit system

In many cases, such as when running proprietary 32-bit games from Steam or in Wine, you may need 32-bit graphics libraries on your 64-bit system in order for them to function properly. This has been made much easier since Debian 9/Stretch and now requires minimal extra work.

Note that the following instructions assume that sudo is configured on your system. If it isn't, either follow the instructions on the sudo wiki page or omit the sudo and run these commands as root.

After installing the drivers, enable 32-bit multiarch and update your repository listing by running:

sudo dpkg --add-architecture i386 && sudo apt update

Afterwards, to install the 32-bit version of the NVIDIA libraries package, run:

sudo apt install nvidia-driver-libs:i386

Restarting the relevant applications may be necessary before they function correctly.

WARNING: If you're forced to use a legacy driver, you will want to instead install one of nvidia-legacy-390xx-driver-libs:i386, nvidia-legacy-340xx-driver-libs:i386, or nvidia-legacy-304xx-driver-libs:i386.


Wayland

The NVIDIA driver supports Wayland, with caveats. The 495-series driver (or newer) is recommended for the best experience, as older versions only support Wayland through an NVIDIA-specific API which is not supported by all desktops, and is generally less reliable.

The NVIDIA driver also lacks support for accelerated XWayland applications in current stable Debian versions. This means that if you run a Xorg-only application on your NVIDIA Wayland desktop (often proprietary video games), they will only be able to render on the CPU without taking advantage of GPU acceleration, leading to incredibly poor performance. Patches have been merged to resolve this, however this support will only be available in Debian 12/Bookworm.

In terms of specific desktop support, GNOME supports NVIDIA Wayland sessions in both Debian 10 and Debian 11, though they call their support "preliminary". KDE Plasma supports NVIDIA Wayland sessions starting with Debian 11, though it requires some extra hoops to enable, and generally is not recommended. Refer to the Wayland section of the Debian KDE wiki page for up-to-date information: https://wiki.debian.org/KDE#Wayland.2C_touchscreens.2C_autorotation.2C_hi-DPI

In Debian 12/bookworm, almost all issues should be resolved and most Wayland sessions should "just work" with the 535-series driver.

On GNOME desktops, although a proper version of the NVIDIA driver is used, the greeter (GDM3) may still not offer the option to start a Wayland session, either because kernel modesetting is not enabled, or because the hibernate/suspend/resume helper scripts aren't enabled or aren't installed on the system.

To enable kernel modesetting with the NVIDIA driver:

# echo 'GRUB_CMDLINE_LINUX="$GRUB_CMDLINE_LINUX nvidia-drm.modeset=1"' > /etc/default/grub.d/nvidia-modeset.cfg
# update-grub

To install the helper scripts, install the nvidia-suspend-common package. Afterwards you need to enable them:

# systemctl enable nvidia-suspend.service
# systemctl enable nvidia-hibernate.service
# systemctl enable nvidia-resume.service

In addition, you will need to verify whether the PreserveVideoMemoryAllocations NVIDIA module parameter is turned on. Without the parameter being enabled, the udev rules in /usr/lib/udev/rules.d/61-gdm.rules will force a fallback to X11. To check the value of PreserveVideoMemoryAllocations:

$ cat /proc/driver/nvidia/params | grep PreserveVideoMemoryAllocations
PreserveVideoMemoryAllocations: 1

If this parameter is set to zero, you should be able to override it by adding a configuration into modprobe.d (assuming the file doesn't already exist):

# echo 'options nvidia NVreg_PreserveVideoMemoryAllocations=1' > /etc/modprobe.d/nvidia-power-management.conf

For more information on PreserveVideoMemoryAllocations: https://download.nvidia.com/XFree86/Linux-x86_64/515.65.01/README/powermanagement.html#PreserveAllVide719f0

Once these changes are made, reboot the system. The GNOME greeter should then start new sessions under Wayland by default.


Tesla Drivers

The NVIDIA line-up of programmable "Tesla" devices, used primarily for simulations and large-scale calculations, also require separate driver packages to function correctly compared to the consumer-grade GeForce GPUs that are instead targeted for desktop and gaming usage.

In Debian 10/Buster, the default nvidia-driver package is based on the Tesla release. This was done in order to resolve several critical security issues, but it means that there is no need to install the separate package for Tesla devices to work. If you need a newer release, the 450-series driver is available in backports via the nvidia-tesla-450-driver package.

In Debian 11/Bullseye, the major 418, 440, and 450 releases of the Tesla driver are available and distinct from the default driver. They can be found in the nvidia-tesla-418-driver, nvidia-tesla-440-driver, and nvidia-tesla-450-driver packages respectively.

In Debian 12/Bookworm, the 470 release of the Tesla driver is available in the nvidia-tesla-470-driver package.

The 32-bit libraries can be obtained by installing nvidia-tesla-418-driver-libs:i386, nvidia-tesla-440-driver:i386, or nvidia-tesla-450-driver:i386 based on the version of your driver. Multiarch must be enabled.


Configuration

As the NVIDIA driver is not autodetected by Xorg, a configuration file is required to be supplied. Modern Debian packages for the NVIDIA driver should not require you to do anything listed here as they handle this automatically during installation, but if you run into issues, or are using a much older version of Debian, you may try going through these steps.

Automatic

Install the nvidia-xconfig package, then run it with sudo. It will automatically generate a Xorg configuration file at /etc/X11/xorg.conf. If the graphics hardware is an eGPU (thunderbolt for example), use the --egpu flag, i.e.: sudo nvidia-xconfig --egpu.

Manual

For example:

/etc/X11/xorg.conf.d/20-nvidia.conf

The configuration file above can be created using these commands:

Please note that this configuration will break Xorg on Optimus systems. For such hardware, see NVIDIA Optimus instead.

Restart your system at this point to enable the nouveau driver blacklist.

Additional configuration information is available.


CUDA

Debian Unstable "Sid"

CUDA 11.8.89 is available from the non-free repository:

# apt install nvidia-cuda-dev nvidia-cuda-toolkit

This installs nvcc and friends. The visual profiler is in a separate package named nvidia-visual-profiler.

Debian 12 "Bookworm"

CUDA 11.8.89 is available from the non-free repository:

# apt install nvidia-cuda-dev nvidia-cuda-toolkit

This installs nvcc and friends. The visual profiler is in a separate package named nvidia-visual-profiler.

Debian 11 "Bullseye"

CUDA 11.2.2 is available from the non-free repository:

# apt install nvidia-cuda-dev nvidia-cuda-toolkit

This installs nvcc and friends. The visual profiler is in a separate package named nvidia-visual-profiler.

Debian 10 "Buster"

CUDA 9.2.148 is available from the non-free repository:

# apt install nvidia-cuda-dev nvidia-cuda-toolkit

And, if Backports are enabled, CUDA 11.2.1 is available similarly:

# apt -t buster-backports install nvidia-cuda-dev nvidia-cuda-toolkit

This installs nvcc and friends. The visual profiler is in a separate package named nvidia-visual-profiler.

Debian 9 "Stretch"

CUDA 8.0.44 is available from the non-free repository:

# apt install nvidia-cuda-dev nvidia-cuda-toolkit

And, if Backports are enabled, CUDA 9.1.85 is available similarly:

# apt -t stretch-backports install nvidia-cuda-dev nvidia-cuda-toolkit

This installs nvcc and friends. The visual profiler is in a separate package named nvidia-visual-profiler.

CUDA 8 only supports gcc 5.3.1, which is not available for Stretch. To compile you need to add -ccbin clang-3.8 to the nvcc command line.

The Debian CUDA packages unfortunately do not include the Toolkit samples. To install these yourself you need to download the "Ubuntu 16.04" .run install file for CUDA 8 from https://developer.nvidia.com/cuda-downloads. Execute the .run file and (after accepting the license and agreeing to run on a non-supported system) skip the driver and toolkit installation and just select "Samples". Note before this step you must

export PERL5LIB=.

To compile the samples, you first need to set

export HOST_COMPILER=clang++-3.8


OptiX

To enable NVIDIA OptiX™ Ray Tracing Engine it is necessary to install an additional library:

# apt install libnvoptix1

A specific version is available for Tesla cards.

# apt install libnvidia-tesla-nvoptix1


Troubleshooting

Build failures

The NVIDIA driver can fail to build for several potential reasons.

1. You've installed a kernel from backports without installing the NVIDIA driver from backports. This can, in some cases, mean that the kernel is too new for the driver version you're attempting to use. Check this by viewing the package description for the NVIDIA driver where it will mention something along the lines of, "Building the kernel module has been tested up to Linux X.X" to figure out what's supported.

2. Particularly if you're on Debian Testing or Debian Unstable, the driver might not support your kernel yet. Often, new versions of the Linux kernel will explicitly require an update to the driver in order to be supported, so if the kernel package updates before the driver has a chance to be patched for it, you won't be able to use the NVIDIA driver. Solutions for this, from most to least recommended, are temporarily using an older kernel until the driver is updated, installing a newer version of the driver from Debian Experimental if one is available that supports your kernel version, or finding a patch for the build failure online that can be added to DKMS. The last two options are for advanced users and may break your system or, in the case of adding a third-party patch, introduce security issues, forcing you to potentially reinstall completely or spend hours recovering your system. Caveat emptor.

3. Legacy versions of the NVIDIA driver may not always support the latest kernel. For instance, the 304xx series driver, though available in the Debian Unstable

repository, does not support Linux 5.0 or newer. As necessary, you might consider using an older Debian version, or using Nouveau instead. Nouveau has decent performance with GPUs that are old enough to no longer be supported by the proprietary driver.

Driver stops working after upgrading Debian

When going between two major Debian releases (e.g., upgrading from Debian 9/Stretch to Debian 10/Buster), it's possible that the driver will stop functioning despite the build succeeding and no other issues being easily visible. This is most often caused by the nvidia-driver package updating to a newer major release that no longer supports your hardware, as NVIDIA regularly drops support for older hardware generations. You will need to uninstall all your existing NVIDIA packages (refer to the section below for instructions on how to do so), and instead install the most recent legacy driver that still supports your GPU.

GPU isn't functional, even with a compatible driver version installed

If you have a hybrid graphics chipset and (after already installing the necessary driver package) 3D acceleration still does not work, you'll still need to choose one of the methods from the NVIDIA Optimus page in order to activate and make use of your NVIDIA card.

If you have an extremely modern NVIDIA GPU that was manufactured after the release of your Debian version, it may not work even after installing the newest backported driver that claims to support your card. If so, you likely need to upgrade the non-free firmware package on your system as well by installing the firmware-misc-nonfree package from backports. For instance, on a Debian 10 system with backports enabled:

# apt install -t buster-backports firmware-misc-nonfree

After rebooting, the driver should be able to load the appropriate firmware.

Miscellaneous


Uninstallation

If you run into issues with the drivers, switch to a different card, or simply want to use the open-source Nouveau drivers instead, uninstallation is made easy with recent versions of the drivers.

Also note that if issues with the driver prevent you from accessing a desktop, you can access a full-screen TTY with Ctrl-Alt-F3 (or almost any of the "F" keys).

You can remove all packages on your system with nvidia in the name by running:

# apt purge "*nvidia*"

And then reboot the system with:

systemctl reboot

This should leave you with a functioning system in almost all cases. If it seems to still be having issues, you may also try running:

# apt install --reinstall xserver-xorg-core xserver-xorg-video-nouveau

Or:

# X -configure

See Also


CategoryProprietarySoftware CategoryHardware CategoryVideo