Ubuntu Version:
25.10
Desktop Environment (if applicable):
KDE Plasma
Problem Description:
Hi all,
I’ve been trying to make cuda toolkit work with ubuntu for llm purposes and made a mess in the process.
Background
I was setting up my ubuntu for llama.cpp using this guide, which requires installing cuda-toolkit from nvidia’s ppa:
Tried it but failed trying to build from source due to a toolchain issue when running cmake (something about incompatible glib and cuda versions requiring naive patch to the math functions).
Only much later did I realize that ubuntu’s official repo already supports cuda-toolkit simply by running apt install nvidia-cuda-toolkit.
Now I have multiple repos for nvidia drivers (ubuntu’s repo and nvidia’s ppa), which is getting confusing. I updated to the latest nvidia driver (595.45.04), thinking that it came from ubuntu’s original repo (it was from nvidia’s repo instead). The update is awesome, but might cause instability down the line.
Question
I want to clean up all the mess I made (remove nvidia repo and drivers I installed through their repo), and return to the original state for easier management.
I’ve been reading around online and people often recommend purging all nvidia drivers (including cuda drivers), removing the nvidia repo, and then reinstalling the drivers using ubuntu-drivers install
Is this the proper solution, or is there a better approach?
Thanks!
Relevant System Information:
Nvidia RTX 5060 Ti (16gb)
Driver: 595.45.04
Cuda: 13.2
Secure Boot Enabled
Screenshots or Error Messages:
N/A
What I’ve Tried:
Considering the recommended approach above, but want to double check to avoid bricking my system.