DigestAI news desk

Cut through the AI noise.

Hardware & Compute3 min read

CachyOS kernel module may unlock RTX 3090, showing 15% performance boost

A Linux kernel module released for the CachyOS 7.2.8-1 distribution combines the CMP unlocker's V67/PLM technique with the pre‑driver timing of the UEFI approach. The module writes speed selectors (SS0 0x88888888 for full BF16 rate and SS1 0x8, whose effect is untested) and performs a fast link reset before the unmodified NVIDIA driver binds the card. In the author’s test, an RTX 3090 at PCI…

1 source primary source

Key points

  • The module unlocks an RTX 3090 on a CachyOS kernel, reporting ~15% pixel and 5% tensor performance gains.
  • Only GA102‑based RTX 3090 cards are supported; other RTX 3090, GA104, or professional GPUs have not been verified.
  • Installation uses make install‑arch/deb/rpm packages and requires a cold power‑off; warm reboots are not supported.

The hack works only on GA102‑based RTX 3090 cards; other RTX 3090 boards, GA104 parts, professional RTX A‑series, or data‑center GPUs have not been verified. Installation is done via make install‑arch, make install‑deb, or make install‑rpm packages that place the module and firmware in the initramfs, loading it before the NVIDIA driver. The unlock requires a cold power‑off; warm reboots are unsupported and will likely fail. All firmware is downloaded over HTTPS, verified with SHA‑256, and the module is GPL‑2.0‑only.

Full story from github.com · via Reddit AI communities primary sourceOpen source ↗

Amazing free 3090 speedup hack, verified works 15% PP, 5% TG for free

github.com · 18 October 2026

Loading the full article…

This text was published by github.com. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗

Coverage and discussion

1source
Topics · follow one to build your own front page
CachyOSNvidiaRTX 3090zagitta

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.

Comments

via GitHub Discussions

More in Hardware & Compute

All →

Related stories