TL;DR
Play games on Amazon Luna for your next adventure, included with Prime
- A rotating selection of games, no download needed
- Play on TV, laptop or phone
- Fast, free delivery for your gear, too
The open-source NV UEFI Unlocker modifies GPU behavior before the NVIDIA driver starts, without permanently flashing the card’s VBIOS. Early tests report up to 75% higher compute throughput on some RTX 20- and 40-series GPUs, but compatibility and real-world gains vary, and RTX 50 cards are not supported.
NV UEFI Unlocker, an open-source utility developed by GitHub user physicsdolphin, is intended to increase Tensor Core compute throughput on supported GeForce RTX 20, 30 and 40 Series GPUs. Early benchmark results cited by Wccftech report gains of up to 75% in a specific matrix-multiplication test, but the project is experimental and those results do not establish a comparable improvement across all AI tasks or systems.
According to the project documentation, the tool applies its modification from a UEFI boot environment, before the NVIDIA graphics driver initializes. Unlike a permanent VBIOS flash or a driver patch, the change is described as temporary and is cleared after the computer is fully powered off. The project documentation says it is designed to work with Windows, Linux and other operating systems that can boot through UEFI.
Wccftech reports that an RTX 2070 reached about 60–63 TFLOPS, compared with a reported 36 TFLOPS baseline, in FP32-accumulation matrix multiplication. That is a stated increase of roughly 67% to 75% for this test. Wccftech also reports that an RTX 4060 Laptop GPU recorded 49.33 TFLOPS in FP16 compute, up from 29.89 TFLOPS, a reported 65% improvement. These are specific test results, not independent confirmation of performance gains on every card.
The project’s installation documentation describes using a bootable USB drive containing the UEFI application and required firmware components. It says Secure Boot must be disabled or configured to trust the EFI application. The documentation also says RTX 20 Series users must enable the required GSP firmware mode. The project’s compatibility matrix lists RTX 20, 30 and 40 Series support, with some model-specific problems; GTX 16 Series support is experimental, and RTX 50 Series cards are unsupported.
GPU COMPUTE · EXPERIMENTAL TOOL
More AI Compute From Your RTX GPU?
NV UEFI Unlocker is an open-source utility that modifies supported GeForce GPU behavior before the NVIDIA driver starts. Early tests report up to 75% higher throughput in selected matrix-multiplication benchmarks—but results vary by card and workload.
A change before the driver loads
The project documentation describes a UEFI-based modification applied during boot. It does not permanently flash the GPU’s VBIOS or patch the NVIDIA driver.
Runs at UEFI boot
A bootable USB contains the UEFI application and required firmware components. The modification takes effect before the graphics driver initializes.
Clears at shutdown
Documentation describes the change as temporary and cleared after the computer is fully powered off, unlike a permanent VBIOS flash.
UEFI-bootable systems
The project says it is designed for Windows, Linux and other operating systems that boot through UEFI.
Two reported results, two workloads
Wccftech cited these figures from specific tests. They do not establish similar gains across all GPUs, AI applications or systems.
RTX 2070
FP32 ACCUMULATIONMATRIX MULTIPLICATION
The upper figure is the source of the “up to 75%” claim.
RTX 4060 Laptop GPU
FP16 COMPUTEBoth values are TFLOPS in the cited FP16 compute test.
Source: Wccftech. These are reported test results, not independent confirmation or a promise of application-level speedups.
What the setup involves
Installation depends on firmware configuration and the specific GPU. Consult the project’s documentation and compatibility matrix before attempting to boot the utility.
Place the UEFI application and required firmware components on a USB drive.
Disable Secure Boot or configure it to trust the EFI application, as the documentation specifies.
RTX 20 Series users must enable the required GSP firmware mode.
The UEFI utility runs before the operating system’s NVIDIA graphics driver initializes.
Generations listed by the project
Support can differ by model. The compatibility matrix includes caveats, and successful operation is not guaranteed for every card.
| GPU family | Project status | What to know |
|---|---|---|
| GeForce RTX 20 Series | Listed | Enable the required GSP firmware mode; model-specific issues may apply. |
| GeForce RTX 30 Series | Listed | Check the project compatibility matrix for model caveats. |
| GeForce RTX 40 Series | Listed | Includes the cited RTX 4060 Laptop GPU test; results vary. |
| GeForce GTX 16 Series | Experimental | Support is experimental. |
| GeForce RTX 50 Series | Unsupported | No support is listed. |
Compute claims need real-world context
Higher theoretical throughput can help selected Tensor Core workloads, but it does not tell the whole performance story.
Local AI experiments
If the reported gains hold on a given setup, local inference, image generation and other matrix-compute tasks may benefit without replacing the GPU.
Workload and system matter
Memory bandwidth, power limits and workload design can constrain results. A synthetic benchmark gain does not predict a proportional reduction in processing time.
Gaming or DLSS gains
The cited reporting does not establish improved gaming performance or faster DLSS execution times.
Broader repeatable testing
Testing across more supported models and real AI workloads—with clear system details and baselines—will help set expectations.
From Linux experiments to UEFI boot
NV UEFI Unlocker follows earlier work on GPU Unlock by Zagitta, which explored higher Tensor Core throughput on RTX 30 Series cards.
Reported tests in selected FP16 and BF16 workloads.
The earlier method relied on a Linux kernel module.
NV UEFI Unlocker applies its modification before the driver starts.
Platform breadth does not remove hardware and firmware caveats.
What users should know
Early figures are a reason to investigate—not a performance guarantee.
Which GPUs are listed?
RTX 20, 30 and 40 Series are listed, with some model-specific issues. GTX 16 support is experimental; RTX 50 is unsupported.
How much faster were the tests?
Wccftech reported about 60–63 vs. 36 TFLOPS for an RTX 2070 FP32-accumulation matrix test, and 49.33 vs. 29.89 TFLOPS for RTX 4060 Laptop FP16 compute.
Does it permanently modify the GPU?
Project documentation describes a temporary UEFI modification rather than a permanent VBIOS flash or driver patch. It is reported to clear after full power-off.
Does it boost gaming or DLSS?
The available reporting does not establish improvements to gaming performance or DLSS execution times.
What setup changes are described?
Boot from a USB with the UEFI app and required firmware components. Secure Boot must be disabled or trust the app; RTX 20 users also need the required GSP mode.
What is the practical takeaway?
A promising, early option for selected compute workloads. Check compatibility and treat reported results as test claims until broader repeatable testing is available.
More Compute From Existing RTX Cards
If the reported results hold for a user’s card and workload, the utility could offer AI and other matrix-compute users a way to raise throughput without replacing an older GPU or permanently changing its firmware. That makes it relevant to people experimenting with local inference, image generation and other applications that use Tensor Cores, particularly where a new graphics card is not an option.
The size of a synthetic throughput gain does not by itself predict a similar improvement in an application. Actual performance can be limited by memory bandwidth, power limits and workload design. The available results also do not show that the modification improves gaming performance or DLSS execution times. For readers, the practical takeaway is a promising but still unverified option for particular compute workloads—not a general performance upgrade.
Products Worth Considering
Amazon Kindle Paperwhite (12th Generation, 2024), 16 GB, 7-inch Display, Jade
Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Black
Anker SOLIX S2000 Portable Power Station, 2,010Wh, 1,500W
As an affiliate, we earn on qualifying purchases.
From Linux Modifications to UEFI Boot
The utility follows earlier work on GPU Unlock, developed by Zagitta. Wccftech previously reported that project’s tests on GeForce RTX 30 Series cards, which showed close to twice normal Tensor Core throughput in certain FP16 and BF16 workloads. Its reported real-world AI image-generation results approached 60% improvement, but the original approach relied on a Linux kernel module.
An experimental, hardware-specific Windows implementation also appeared around that time. NV UEFI Unlocker takes a different route, running before the operating system’s graphics driver and listing compatibility across three RTX generations. The broader platform support does not remove the project’s experimental status: the hardware matrix includes caveats, and each system’s firmware configuration can affect whether the tool works.
Compatibility and Real-World Gains
The reported benchmarks do not establish how often the stated gains can be reproduced, which precise GPU configurations were used beyond the listed examples, or how performance changes across different AI software and models. Not every supported card is guaranteed to unlock successfully, and the project’s compatibility matrix documents issues with some models. GTX 16 Series support remains experimental, while RTX 50 Series support is absent.
It is also unclear whether a given workload will benefit once other system limits are taken into account. Higher theoretical compute throughput may not translate into a proportionate reduction in processing time, and the results cited by Wccftech do not demonstrate improvements to games or DLSS. Users should treat the figures as early test claims rather than a general performance guarantee.
Further Testing Will Set Expectations
The next useful milestone is broader testing across supported GPU models and real AI workloads, with clear reporting of system configuration, baseline methods and repeatability. Users considering the tool can consult the project’s official compatibility matrix and installation documentation before attempting to boot it, especially if Secure Boot or GSP settings need to change.
No release schedule for a more mature version or RTX 50 Series support is established in the available reporting. Until further results are published, the project’s reach and practical performance remain developing questions. The figures reported by Wccftech show what early tests have achieved, not what every owner should expect.
Source: Wccftech
Key Questions
Which NVIDIA graphics cards does NV UEFI Unlocker support?
The project’s compatibility matrix lists RTX 20, 30 and 40 Series cards, though some models have issues. GTX 16 Series support is experimental, and RTX 50 Series GPUs are not supported.
How much faster did the reported tests get?
Wccftech reported an RTX 2070 result of about 60–63 TFLOPS versus 36 TFLOPS in FP32-accumulation matrix multiplication, and an RTX 4060 Laptop GPU result of 49.33 versus 29.89 TFLOPS in FP16 compute. The figures apply to those tests and do not promise equivalent gains in other workloads.
Does the tool permanently modify the GPU?
The project documentation describes a temporary UEFI-based modification, rather than permanently flashing the VBIOS or patching the NVIDIA driver. The change is reported to clear after a complete power-off.
Does NV UEFI Unlocker improve gaming or DLSS performance?
The available reporting provides no evidence establishing that it improves gaming performance or DLSS execution times. Its reported gains concern compute benchmarks and selected AI-related workloads.
What system changes does installation require?
The project’s installation documentation describes booting a USB drive containing the UEFI application and required firmware components. Secure Boot must be disabled or configured to trust the EFI application; RTX 20 Series users also need to enable the required GSP firmware mode.
Source: Wccftech
