Here’s the news stand for the nvidia-inst app.
nvidia-inst is a driver installer for Nvidia GPUs, and it installs packages only from the Arch repos.
If you need other Nvidia drivers from the AUR, you should install them manually.
The news
Release 25.4.3-1
This latest release of package nvidia-inst has support for automatically selecting the Nvidia open source driver (nvidia-open-dkms and related) and Nvidia closed source driver (.nvidia-open-dkms and related)
This means you’ll automatically get the GPU driver recommended by Nvidia.
Note that new GPUs should use the open source version, and slightly older need the closed source version.
And if you want to control which one to choose, use options --open or
--closed.
Please report if you experience any issue with the latest version. ![]()
Note also that a yesterday’s bugfix corrects the behavior of option --prime.
More info with command: nvidia-inst --help
Release 25.4.4-1
Added new option --show-gpu-info which shows some essential information about your NVIDIA GPU(s). It gives very useful information for improving
nvidia-inst with various NVIDIA GPU families and troubleshooting their issues.
Release 25.4.5-1
Since nvidia-inst by default favors the dkms package versions, this release also makes sure the required kernel headers and the dkms package will be installed. To use the non-dkms packages, add option --no-dkms to nvidia-inst.
Releases 25.4.6-1 - 25.4.11-1
Notable improvements:
- assume option
--32when[multilib]repo is enabled in/etc/pacman.conf, andlib32-nvidia-utilsis installed
- minor output fixes and code cleanup
Release 25.12-1
This release takes into account the new Nvidia and removed packages in the Arch repos. Now nvidia-inst will install
nvidia-open-dkms(ornvidia-open) package for Turing and newer GPUsvulkan-nouveaufor older Nvidia GPUs
Note that for Maxwell, Pascal, and Volta GPUs nvidia-inst will inform user about the AUR package nvidia-580xx-dkms if this driver supports the installed GPU.
Also, nvidia-inst will ask a confirmation if user accidentally uses --open with a GPU older than Turing.