Back

Open-source USB tool unlocks Nvidia's half-speed GPU AI limiter on Windows
SiTech AI Team2 min read

Open-source USB tool unlocks Nvidia's half-speed GPU AI limiter on Windows

A new open-source UEFI unlocker boots from USB, unlocks Nvidia's tensor limiter on Windows for RTX 20, 30, and 40-series GPUs, and reportedly lifts FP16 performance by up to 65% in one developer test.

A new open-source UEFI tool boots from USB and unlocks Nvidia's tensor limiter on Windows, aiming at RTX 20, 30 and 40-series desktop and laptop GPUs. The nv-uefi-unlocker project can raise FP16 math performance, with the developer's own laptop RTX 4060 reaching 65% higher FP16 math in one test, without flashing the video BIOS or modifying the driver.

How the unlock works

The tool removes Nvidia's tensor limiter using the community V67 payload with read-back verification. It tricks Nvidia's signed Booter firmware into opening the write lock on two speed-selector registers, allowing tensor units to run at full rate. It runs from a FAT32 USB stick before the Nvidia driver loads at every boot, so the unlock does not persist through power loss and must be repeated each boot.

Setup and compatibility

The release includes scripts and self-built binaries, but no Nvidia firmware. One script downloads Nvidia's public 610.43.03 driver packages, verifies them, and rebuilds firmware files locally. The system must boot in UEFI mode with Secure Boot off or set to trust a self-signed EFI; the tool cannot bypass Secure Boot for game anti-cheat. RTX 20 owners must run a bundled script once to switch the Windows Nvidia driver into GSP firmware mode, then restart. Laptop GPUs need the discrete Nvidia GPU powered at entry, so dGPU-only or MUX mode is recommended.

Reported results and limits

Reported Windows results include an RTX 2070 rising from 36 to 60-63 TFLOPS in FP16 matrix multiply, a 65% gain on an RTX 4060 laptop, and an RTX 4080 rising from 100.37 to 192.18 TFLOPS in BF16 matrix multiply. Compatibility lists RTX 20-series as fully compatible with some known issues, RTX 30-series as fully compatible, and RTX 40-series as fully compatible on laptops and desktops. GTX 16 is theoretically compatible with verification pending, while pre-Turing, GA100 and RTX 50-series cards are unsupported.

Caveats

Results depend on power and thermal limits, and combining the unlock with the nvoc power-limit unlock can exceed factory power and thermal envelopes. A per-card ROM hash check required by the project's safety rules is not built yet, Thunderbolt eGPU persistence is not guaranteed, and cards such as the RTX 2080 and RTX 2080 Ti are under review. RTX 50-series cards have more security than anticipated, and the team says a boot-from-disk toggle is in the works.

Sources: Tom's Hardware

SiTech

SiTech — AI-powered web development

We build fast, modern websites and bring AI into real business workflows. Have a project or a question? We'd love to help.