gpu-unlock

A Linux kernel module that unlocks FP/BF16 instruction issue rates on supported Nvidia RTX 30-series cards. It is used to improve matrix-computation throughput without patching the Nvidia driver.

Browse open issues

Pick a small issue you can reproduce, then fix it with your AI coding agent.

Share on XNo license declared

Overview

GPU unlock is a Linux kernel module for raising FP/BF16 instruction issue rates on supported Nvidia RTX 30-series cards. It performs the unlock before the unmodified Nvidia driver binds, writes the speed selectors, and runs a function-level reset. The project reports doubled FP/BF16 FMA issue speed for FP32 operations on GA102, GA103, GA104, and GA106 chips. Users apply it to improve matrix-computation throughput, though reported results vary by card and power limit.

Key features

  • Runs before the Nvidia driver binds and leaves the driver unmodified.
  • Writes speed selectors and performs a function-level reset as part of the unlock.
  • Targets FP/BF16 FMA issue speed for FP32 operations on GA102, GA103, GA104, and GA106 chips.
  • Includes user-reported compatibility results for several RTX 30-series cards.

Best for

Linux users with a supported RTX 30-series card who want higher matrix-computation throughput without patching the Nvidia driver. Check the reported compatibility for the specific card, since results vary and some models lack reports.

Upstream
Zagitta/gpu-unlock
Fork on GitHub
Guo-astro/gpu-unlock
Upstream stars
3.2k
Category
Hardware, OS and embedded
Language
C
License
No license declaredWithout a license, the author keeps all rights. Ask the upstream owner before reusing the code.
Forked
2026-10-24
Sync status
Updated by last syncLast synced 2026-10-25