gitmyhub

nvidia-patch

Python ★ 4.7k updated 22h ago

This patch removes restriction on maximum number of simultaneous NVENC video encoding sessions imposed by Nvidia to consumer-grade GPUs.

Two shell scripts that patch Nvidia's Linux drivers to remove the artificial limit on simultaneous video encoding sessions, unlocking the GPU's full hardware capabilities for consumer cards.

PythonShellsetup: easycomplexity 2/5

Nvidia sells graphics cards aimed at consumers and a separate line aimed at professionals and data centers. One difference between them is a software-imposed limit: consumer cards are restricted to a small number of simultaneous video encoding sessions, while the professional cards are not. This limit is enforced inside the Nvidia driver software, not in the hardware itself.

This project provides two shell scripts that modify the Nvidia driver files on a Linux system to remove those restrictions. The first script, patch.sh, removes the cap on how many simultaneous NVENC encoding sessions you can run. NVENC is the hardware video encoding engine built into most Nvidia cards, used by software like OBS, Handbrake, Plex, and others. The second script, patch-fbc.sh, unlocks NvFBC, a feature that lets software capture the screen through the GPU rather than through slower software methods.

The README includes a version table listing every supported Nvidia driver version, with direct download links for each. You check your installed driver version, find it in the table, run the appropriate patch script, and the restriction is gone. The project notes which driver versions support just the NVENC patch, and which also support the NvFBC patch. The primary target is GNU/Linux, though there is a separate folder with notes for Windows users.

This patch is useful for anyone running a media server, a streaming setup, or video transcoding work on a consumer Nvidia card who keeps running into session limit errors. Without the patch, you might hit the cap at just two or three simultaneous encoding jobs. With the patch removed, the only ceiling is what the GPU hardware itself can actually handle.

The project is open source and MIT licensed. The author asks that donations go to open source organizations like FFmpeg or VideoLAN rather than to the project directly. The full README is longer than what was shown.

Where it fits