Nvidia Linux one-click auto undervolt and Afterburner profile import is now possible
This seems to finally be the Linux Gaming/Desktop year and happy to see that Nvidia is also upping the Linux game with Proton heap patches, new drivers, generally lots of quality of life improvements this year!
The main missing thing for my RTX 5080 Linux setup was sort of a one-click undervolting tool, similar to NV-UV on Windows, which was linked here a month ago:
https://www.reddit.com/r/nvidia/comments/1ro93n0/nvuv_brings_oneclick_undervolting_to_geforce_rtx/
Also wanted to have an easy way to import Afterburner VF curve profile (as is) to Linux.
Unfortunately the NVUV does not have any GPL source code on Github available, just the release with a windows binary file. So I decided to team up with Codex and solved it with a new tool, you can take a look at the link below below.
Install through Fedora COPR, Ubuntu PPA, Arch AUR or just python pip install.
https://github.com/jpietek/PenguinBurner
For all the anti AI Reddit community: we can like it or not but LLMs generate vast portion of our graphics drivers now too.
But seriously Codex+GPT 5.5 xhigh was pretty helpful with reverse engineering parts and was able to gather a lot of instability cases by trial and error with feedback from hundreds of Auto UV runs that I did. For stability verification I use a mix of offscreen Q2RTX path traced loop and custom CUDA compute accuracy test.
Obviously the code readability takes a big hit, but the whole algorithm is specified and clear, everything open source - autouv3 dir, after third refactor to make is as human readable as possible. You can also read the Auto UV procedure in plain text - step by step, generally it should promote the best efficiency, so FPS per Watt.
https://github.com/jpietek/PenguinBurner/issues/6
On my personal 5080 went down -21% with power and +1.8% with FPS in performance Auto UV mode, so not bad I think.
The tool was already tested with dozen or 30/40/50 series cards and worked pretty well, the bugs ironed out, so you don't have to worry as the alpha early stage testing is already done.
Have fun, and happy to hear any (possibly) constructive feedback.
