NV UEFI Unlocker Boosts AI Speed on RTX 20, 30 and 40 GPUs

NV UEFI Unlocker, a new open-source tool from GitHub developer physicsdolphin, claims to lift the Tensor core limits NVIDIA places on GeForce cards and raise AI compute on RTX 20, RTX 30 and RTX 40 series GPUs by up to about 75 percent in early tests. Version 2.1.8 was released on October 31, and unlike earlier Linux-only hacks, it also works on Windows.

The trick is timing. The tool runs as a UEFI application from a USB stick before the operating system and NVIDIA’s driver load, unlocks the card, then hands control back to Windows, Linux or another EFI-bootable system. No VBIOS flashing or driver patching is needed, and a full power-off clears the unlock, so it has to be repeated after every cold boot.

How Much Faster Is NV UEFI Unlocker in AI Workloads?

The project’s GitHub page lists measured gains of up to about 2.6 times in FP16, BF16 and TF32 tensor throughput, up to 1.9 times in FP8, 14 to 75 percent in image and video generation and up to 94 percent in DirectML inference kernels. The developer notes that power limits and precision paths cap the real gains.

Early user results point the same way. An RTX 2070 reportedly went from 36 TFLOPS to about 60 to 63 TFLOPS in matrix multiplication with FP32 accumulation, a 67 to 75 percent jump, while an RTX 4060 Laptop GPU climbed from 29.89 to 49.33 TFLOPS in FP16, according to Wccftech.

Which GPUs Are Supported?

• RTX 40 (Ada): supported on desktops and laptops.
• RTX 30 (Ampere GA10x): supported.
• RTX 20 (Turing): supported, but some models, including certain RTX 2080 and 2080 Ti boards, have known issues.
• GTX 16: theoretically compatible, not yet verified.
• RTX 50, Pascal and older, and GA100: not supported.

Setup needs a FAT32 USB stick, UEFI boot, and Secure Boot either disabled or configured to trust the EFI file. RTX 20 owners must first enable NVIDIA’s GSP firmware mode on Windows. If you are unsure what turning off Secure Boot means for your PC, our guide to Windows 11 Secure Boot certificates explains why it matters.

What It Means for Gamers and Local AI Users

Do not expect higher frame rates. There is no convincing evidence that the unlock improves gaming or DLSS performance, since the gains show up mainly in AI workloads. The project also warns that pairing it with overclocking tools can push power and temperatures beyond factory limits, at the user’s own risk.

For people running local language models or image generators on older cards, the appeal is clear while memory prices keep rising, a trend Sam Altman recently linked to AI data centers. It is also another sign of how far the GPU modding community is going, after claims of RTX 5090 cards modded to 96GB. NVIDIA has not commented on the tool, and the project remains experimental, so results will vary by card and system.