GPU Unlock
Zagitta/gpu-unlockReviewed by AI
Radar found strong risk signals in this project. It stays visible so you can read the evidence, but it is left out of rankings and recommendations.
What it does
A Linux kernel module that loads at startup, before the NVIDIA driver, and removes the cap NVIDIA puts on GeForce tensor cores for FP16, BF16 and TF32 maths with FP32 accumulate. The README reports about double the matrix-multiply rate (RTX 3090: 75 to 125 TFLOPS) and, from user reports, 25–55% faster LLM prompt processing and image generation, with no change for gaming, quantized models or token generation. RTX 30 cards are called stable; RTX 20 and 40 are experimental and RTX 50 is not supported. You build it with make install, add a kernel boot option and restart.
How easy it is to install
E · developers only · Linux
Linux is the only system it supports and there it is an E: you clone the repository, run make install (it builds a kernel module and installs distribution packages with your sudo password), then edit the kernel command line in GRUB, systemd-boot or your UKI setup by hand and restart. The author says he does not run Windows and only points at someone else's port.
Coming soon: install in 1 click with the RepoPicks app.
Numbers on GitHub
- Stars
- 13,854on Oct 31, 2026
- Downloads
- 447447 this week
- Growth
- +4,123stars in 5 days
Stars over time
Facts
- Runs on
- Linux
- You need to install
- NVIDIA graphics card, Other tools
- Disk space
- not determined
- Accounts or keys
- None
- Cost
- Free
- License
- No license (No license)
- Last update
- Oct 28, 2026
Risk signals (8)
- Highvendor firmware lock bypass
This is the project's purpose. README: it "combines the cmpunlocker's V67/PLM technique with the pre-driver timing of the UEFI approach" and writes the speed selectors "SS0 0x88888888" and "SS1 0x8" before the NVIDIA driver binds; the FAQ says NVIDIA capped the cards for "Market segmentation". Removing a limit the maker built into the card goes against NVIDIA's terms and can count as circumventing a technical protection, which is restricted in some countries.
AI review
- AttentionVery new repository with many stars
11390 stars in 10 days
automatic check
- Attentionfake installers and coins circulating
FAQ: "Do you have a website/coin/installer? A: No this is the one and only OG source for gpu-unlock. Crypto, websites, installers and what else is not associated with this project or the developer in any way shape or form." And about Windows: "There are windows installer's floating around but I would recommend being ++very++ careful with those."
AI review
- Attentionhardware instability risk
README: "If the firmware on a card never stops, the module logs it and keeps the DMA buffers that firmware may still use until the next power off. Power off before trying again." The display card's "screen is blank from that reset until nvidia-drm takes over". FAQ: "It may heat up your GPU slightly".
AI review
- InfoNo license detected
No license detected by GitHub, the license file, the LICENSES folder or the winget manifest
automatic check
- Infolicense stated only in readme
No license file was detected, but the README says "The kernel module and local packaging are GPL-2.0-only" and points to a licenses/ folder; "NVIDIA firmware remains under its own terms".
AI review
- Infobuild downloads nvidia files
README, Local build: "Every build downloads the NVIDIA/CachyOS inputs over HTTPS and generates firmware only in the build directory and the binary package." Packages are "built from the working tree, uncommitted changes included".
AI review
- Infocrypto donation addresses
README "Support me" lists ETH, BTC and SOL wallet addresses alongside GitHub Sponsors and Buy Me a Coffee.
AI review
Evidence is shown in its original language.
What Radar could not check
- Whether it works on your card, motherboard and Linux distribution; the figures come from users' own reports in the issues
- Long-term stability, heat and power draw after the change
- How much of the doubled rate reaches the AI programs you actually use
- Whether the unlock holds on a laptop, since the README says powering the GPU down "silently resets the unlock"
- What the generated firmware buffer, derived from "the pinned CMP reference", does beyond the README's description
- Warranty and legal consequences of changing NVIDIA's locked settings in your country