Hello,
I recently bought a new RTX 4060 Ti 16GB and installed a new clean Debian 12 with the nvidia-driver from the Debian repositories. According to this link, the RTX 4060 Ti should be supported by the driver version 535.216.01 which was installed. Also nvidia-detect came to the same result. To install, I followed this tutorial. Which means I installed linux headers, added contrib and non-free repositories and then installed the driver via apt.
After a reboot, the boot process got stuck with a blinking underscore. Opening a terminal and running the command echo "options nvidia-drm modeset=1" >> /etc/modprobe.d/nvidia-options.conf as described in the tutorial at least helped to get a desktop again. However, the resolution was still bad and the GPU is still not properly working. Running nvidia-smi results in “No devices were found”. In nvidia-settings, I cannot see the GPU and any settings. I attached the nvidia-bug-report for this issue.
For testing purposes, I re-installed a clean Debian again and installed the current driver version from the Nvidia website. The installation worked and the driver seems to work, as I immediately get the correct resolution and also the nvidia-settings works as expected. So I don’t think it’s a hardware issue. However, I prefer the way via the official Debian repositories to avoid issues with future system or kernel updates. Since the GPU should be supported also by the older driver version in the Debian repository, I wonder what the issue with that is?
Can anybody help understanding, why the driver from repository is not working? Can I fix it somehow, so that I can avoid the manual installation and maintenance later?
nvidia-bug-report.log.gz (162.1 KB)
1 Like
Have you tried version 535.216.03 from the official Debain repos? FWIW, It used to work fine with 3090. (it may come from testing series: I’m on Trixie).
Again FWIW, I’ve been using Nvidia’s apt Debian repo for some time now and never had any problems with updates (both in case of driver updates by Nvidia and kernel updates by Debian). Not sure if it was necessary, but i remember that before switching to Nvidia’s repo I’ve run sudo apt-get purge '*nvidia*' '*nvidia*:i386' '*cuda*:i386' '*cuda*' libxnvctrl0:i386 libxnvctrl0 just in case. After adding Nvidia’s repo in apt config, I installed package nvidia-open.
Yes, I tried from official Debian repo. To be honest, the Nvidia repo I didn’t know so far, usually I found tutorials to use either Debian repo or the .run file from Nvidia website.
Is that Nvidia repo you mentioned something official, meaning it contains the official versions which are also available on the website? Or are the versions there some experimental or development versions, which are not recommended for daily use? At least I didn’t find any official documentation yet, which recommends using this repo.
The Nvidia apt repo is a “compute” or “datacenter” repo, so in theory it should be rock-stable at least for compute stuff. It does contain all the desktop stuff as well nevertheless. According to this post, within the same versions, files in the deb packages and .run executables are exactly the same. The latest versions available from the network apt repo usually correspond to beta versions of drivers from .run files, however this repo keeps older versions as well so you can choose which one to install (currently going back as far as 545).
If you go to datacenter-driver Downloads | NVIDIA Developer and select “Linux” → “x86_64” → “Debian” → “12” → “deb (network)”, you will get instructions to download a deb from exactly the repo I posted. Here is the official documentation: 1. Introduction — NVIDIA Driver Installation Guide r570 documentation