Graphics processing unit

A graphics processing unit (GPU) is the hardware in your computer that generates the video feed that appears on your display. They are present in two device types: Integrated Graphics Processors (IGP) and dedicated (or discrete) graphics, also known as graphics cards. The acronym is often abused to refer to the graphics card as a whole.

Their ease to perform parallel calculation has given birth to general-purpose computing on graphics processing units (GPGPU).

Installation

The Linux kernel includes open-source video drivers and support for hardware accelerated framebuffers. However, userland support is required for OpenGL, Vulkan, 2D acceleration in Xorg and hardware video acceleration.

First, identify the graphics card (the Subsystem output shows the specific model):

$ lspci -vnnd ::03xx

Tip ::03 here means "Display controller PCI device class", and xx stands for "any subclass of the class".

Then, installing its video driver is required. See the tables below for the three major vendors.

For 32-bit software, enable the multilib repository and install lib32- prefixed userspace drivers, such as lib32-mesa, lib32-vulkan-radeon, lib32-nvidia-utils, etc.

For X11, other Device Dependent X (DDX) drivers can be found in the xorg-drivers group or searching for xf86-video. In most cases, you do not need to install any DDX drivers; see Xorg#Drivers for details.

AMD

AMD supports the open source driver. A proprietary driver was provided before but it is no longer packaged. [1][2]

Tip For a translation of model names (e.g. Radeon RX 6600) to GPU family (e.g. RDNA 2), see Wikipedia:List of AMD graphics processing units#Features overview.

See Hardware video acceleration#Comparison tables for details on VA-API support per GPU family.

Documentation	GPU family	DRM driver	OpenGL	Vulkan	DDX driver
AMDGPU	GCN 3 and later (e.g. RDNA 1-4)	included in Linux	mesa	vulkan-radeon	xf86-video-amdgpu²
AMDGPU¹ / ATI	GCN 1&2			Depends on the chosen driver
ATI	R300 through TeraScale			None	xf86-video-ati²
	R100 & R200		mesa-amber	None	xf86-video-ati²
	Rage 4 and older	not available [3][4]

Enabled by default since linux≥6.19, can be manually chosen otherwise
Using the modesetting driver is reported to work without issues

Intel

Intel provides and supports open source drivers, except for PowerVR-based graphics (GMA 3600 series) which are not supported.

Tip Intel's Gen N hardware does not refer to the generation of the CPU, it refers to the generation of the GPU, which is different from the generation of the CPU.

See Hardware video acceleration#Comparison tables for more details on VA-API support per GPU family, only the packages are listed below.

Documentation	GPU family	DRM driver	OpenGL	Vulkan	DDX driver	VA-API
Intel graphics	Gen 12.1 and later	included in Linux	mesa	vulkan-intel¹	xf86-video-intel²	intel-media-driver
	Gen 8 through 11					intel-media-driver or legacy libva-intel-driver
	Gen 7 & 7.5					libva-intel-driver
	Gen 5 & 6			None		libva-intel-driver
	Gen 3 through 4.5					None
	Gen 2		mesa-amber			None

Gen 7 and 7.5 have incomplete support, Gen 8 is limited to Vulkan 1.3.
The modesetting DDX driver is recommended for Gen 3 hardware and later. See Intel graphics#Installation for details.

NVIDIA

NVIDIA does not support the fully open driver. They switched to a hybrid approach (with an open DRM driver and closed userland) in 2022.

Tip For a translation of model names (e.g. RTX 4060) to GPU family (e.g. NV190), see nouveau wiki's code names page.

See Hardware video acceleration#NVIDIA for details on which APIs are supported, and Hardware video acceleration#VDPAU drivers for details per GPU family.

License	Documentation	GPU family	DRM driver	OpenGL	Vulkan	DDX driver
Open	Nouveau¹	Kepler (NVE0/GKXXX) and newer	included in Linux	mesa	vulkan-nouveau	xf86-video-nouveau²
Open	Nouveau¹	Fahrenheit (NV04/05) through Fermi (NVC0/GF1XX)	included in Linux	mesa	None	xf86-video-nouveau²
Open DRM driver, proprietary userland	NVIDIA¹	Turing (NV160/TUXXX) and newer	nvidia-open	nvidia-utils
Proprietary		Maxwell (NV110/GMXXX) through Ada Lovelace (NV190/ADXXX)	nvidia-580xx-dkms^AUR	nvidia-580xx-utils^AUR
		Kepler (NVE0/GKXXX)	nvidia-470xx-dkms^AUR	nvidia-470xx-utils^AUR
		Fermi (NVC0/GF1XX)	nvidia-390xx-dkms^AUR	nvidia-390xx-utils^AUR	None	nvidia-390xx-utils^AUR
		Tesla (NV50/G80-90-GT2XX)	nvidia-340xx-dkms^AUR	nvidia-340xx-utils^AUR	None	nvidia-340xx-utils^AUR
		Curie (NV40/G70) and older	No longer packaged

For NVIDIA Optimus enabled laptop which uses an integrated video card combined with a dedicated GPU, see NVIDIA Optimus.
The modesetting DDX driver is recommended for for NV50 (G80) and later. See Nouveau#Installation for details.

Loading

Most driver kernel modules should load automatically on system boot.

If it does not happen, then:

Make sure you do not have nomodeset or vga= as a kernel parameter, since they require KMS.
Also, check that you have not disabled the driver by using any kernel module blacklisting.

Monitoring

Monitoring your GPU is often used to check the temperature, core and VRAM utilization, and the P-states of your GPU.

CLI

amdgpu_top — Tool to display AMDGPU usage

https://github.com/Umio-Yasuno/amdgpu_top || amdgpu_top

nvtop — GPUs process monitoring for AMD, Intel and NVIDIA

https://github.com/Syllo/nvtop || nvtop

radeontop — A GPU utilization viewer, both for the total activity percent and individual blocks. Works with ATI's R600 and newer GPUs as well as cards using the AMDGPU driver

https://github.com/clbr/radeontop || radeontop