New NVIDIA UEFI GPU Unlocker Brings Up To 75% Higher AI Compute Performance To RTX 20, 30 & 40 Series GPUs On Windows
AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

The open-source NV UEFI Unlocker modifies GPU behavior before the NVIDIA driver loads and reports higher AI compute throughput on some GeForce RTX 20, 30 and 40 Series cards. The project is experimental, its benchmark gains are not guaranteed across systems, and there is no convincing evidence that it improves gaming performance or DLSS execution times.

The open-source NV UEFI Unlocker is reporting higher AI compute throughput on some NVIDIA GeForce RTX 20, 30 and 40 Series graphics cards, using a temporary UEFI-stage modification that can run before Windows or Linux loads. Early tests cited for the project show gains of up to 75% in a specific matrix-multiplication benchmark, but results vary by GPU and workload, and the tool is not a guaranteed performance upgrade.

For playersOffer from Amazon

Play games on Amazon Luna for your game nights, included with Prime

  • A rotating selection of games, no download needed
  • Play on TV, laptop or phone
  • Fast, free delivery for your gear, too
Start playing with Prime Free trial · Cancel anytime
As an affiliate, we earn on qualifying purchases.

Developed by GitHub user physicsdolphin, the utility changes GPU behavior before NVIDIA’s graphics driver initializes. Its UEFI-based method does not require users to permanently flash the card’s VBIOS or patch NVIDIA’s driver. The modification is temporary and is removed after a complete power-off, according to the project description. It is designed to work with Windows, Linux and other operating systems that can boot through UEFI.

Reported tests include an RTX 2070 reaching roughly 60–63 TFLOPS from a stated baseline of 36 TFLOPS in FP32-accumulation matrix multiplication, an increase of about 67% to 75% in that test. An RTX 4060 Laptop GPU reportedly rose from 29.89 to 49.33 TFLOPS in FP16 compute throughput, or about 65%. These are early, workload-specific results; they do not establish similar gains across all applications or graphics cards.

The project’s compatibility information lists RTX 20, 30 and 40 Series support, while noting that some models have issues. GTX 16 Series support is experimental, and RTX 50 Series cards are not supported. The documented setup involves preparing a bootable USB drive with the UEFI application and required firmware components, then booting from it so the utility can run before handing control to the operating system. 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.

At a glance
reportWhen: Project and early benchmark results are…
The developmentAn open-source UEFI utility has reported substantial AI compute throughput gains on selected NVIDIA GeForce RTX cards, with support for Windows and Linux.
Top Steam deals right now
Sons Of The Forest-72%$8.39
Resident Evil Requiem-30%$48.99
Gothic 1 Remake-25%$37.49
The Wolf Among Us Remastered-20%$19.99
Terrible Lizards-20%$19.99
Queen’s Domain-15%$16.99
CRACK-15%$6.79
Probably Stolen – Cyberpunk Shopkeeper Sim-10%$17.99
Live · Steam store (current discounts)
What the NVIDIA UEFI GPU Unlocker Actually Shows

GPU COMPUTE · EARLY PROJECT REPORT

More AI Compute From Some RTX Cards?

The open-source NV UEFI Unlocker reports higher AI compute throughput on selected GeForce RTX 20, 30 and 40 Series GPUs. Early benchmark gains are workload-specific, experimental and not guaranteed.

20–40RTX series listed
+67–75%RTX 2070 test report
+65%RTX 4060 laptop test
0 proofGaming or DLSS gains

01 / WHAT IT DOES

A temporary change before the driver loads

The utility runs through UEFI before the NVIDIA graphics driver initializes. The project’s description says it does not require a permanent VBIOS flash or driver patch, and that its modification disappears after a complete power-off.

MECHANISM

UEFI stage

The tool changes GPU behavior early in the boot process, before Windows or Linux hands control to the graphics driver.

SCOPE

Compute focused

Reported gains concern particular AI compute workloads and throughput measurements, not a broad performance guarantee.

DURATION

Power-off reset

Project documentation describes the change as temporary and says it is removed after a complete power-off.

02 / REPORTED BENCHMARKS

Big numbers, narrow tests

These early figures show results from specific cards and compute formats. They should be read as project reports, not expected outcomes for every model or application.

GPUTestReported resultChange
RTX 2070FP32-accumulation matrix multiplication36 → 60–63 TFLOPS+67–75%
RTX 4060 Laptop GPUFP16 compute29.89 → 49.33 TFLOPS~+65%
Reported baseline29.89 TFLOPS
Reported with unlocker49.33 TFLOPS

03 / DOCUMENTED BOOT PATH

How the setup is described

The project instructions involve a bootable USB setup and model-specific configuration. Requirements may vary by card; check current project documentation before attempting it.

01

Prepare USB

Place the UEFI application and required firmware components on a bootable drive.

02

Check Secure Boot

Disable Secure Boot or configure the system to trust the EFI application.

03

Apply model settings

RTX 20 Series users also need to enable the required GSP firmware mode.

04

Boot the utility

Run it before the system hands control to the operating system.

Experimental means uncertain

Successful unlocking is not assured. The project lists some model-specific issues, and installation depends on firmware and boot settings. Follow the current hardware-specific instructions carefully.

04 / COMPATIBILITY SNAPSHOT

Support varies by generation

LISTED SUPPORT

RTX 20, 30 & 40 Series

These generations are listed as supported, though some models reportedly have issues. Confirm the compatibility matrix for the exact card.

LIMITS

GTX 16 experimental

GTX 16 Series support is experimental. RTX 50 Series cards are outside the stated supported range.

05 / WHAT THE DATA DOES NOT SHOW

Compute throughput is not application speed

Synthetic or narrowly defined TFLOPS measurements do not directly predict task completion time. Real performance depends on the workload, memory bandwidth, power limits and system configuration.

POSSIBLE USE

Local AI tasks

Higher throughput could matter for some local image-generation or inference workloads if results hold for that card and task.

NOT ESTABLISHED

Gaming performance

Available reporting provides no convincing evidence that the modification improves game performance.

NOT ESTABLISHED

DLSS execution

The cited benchmarks do not demonstrate faster DLSS execution times or gains across graphics applications.

“The modification is temporary and disappears after a complete power-off.”

NV UEFI Unlocker project documentation

06 / KEY QUESTIONS

What to know before drawing conclusions

Which GPUs are listed?

GeForce RTX 20, 30 and 40 Series. Some models have issues; GTX 16 Series support is experimental, and RTX 50 Series is unsupported.

Does it permanently change the VBIOS?

The described method runs through UEFI before driver initialization and does not require a permanent VBIOS flash or NVIDIA driver patch.

Are the benchmark gains guaranteed?

No. The numbers are early, workload-specific reports and do not establish results for every GPU, system or application.

Will gaming or DLSS get faster?

There is no convincing evidence in the available reporting that gaming performance or DLSS execution times improve.

What does setup involve?

A bootable USB drive with the UEFI application and required firmware components, plus Secure Boot configuration. RTX 20 Series also requires the specified GSP firmware mode.

What evidence is needed next?

Independent tests across more supported cards, operating systems and real AI workloads would clarify reliability and practical task-time gains.

More AI Compute From Existing RTX Cards

If the reported results hold for a user’s card and intended workload, the utility could let owners test higher AI compute throughput without buying a newer GPU or making a permanent VBIOS change. That may be relevant to people running local AI image-generation or inference workloads, where access to compute capacity can affect processing time.

However, TFLOPS figures from a synthetic or narrowly defined test do not directly predict application-level improvements. Real results depend on the specific workload, memory bandwidth and the card’s power limits, among other factors. The available information also provides no convincing evidence of gains in gaming performance or faster DLSS execution. Readers should treat the numbers as reports of particular tests, not a promise of a broad performance increase.

KitchenAid Classic Series 4.5-Quart Tilt-Head Stand Mixer, Onyx Black

KitchenAid Classic Series 4.5-Quart Tilt-Head Stand Mixer, Onyx Black

De'Longhi Classic Signature Espresso Machine with Adjustable Milk Frother

De'Longhi Classic Signature Espresso Machine with Adjustable Milk Frother

As an affiliate, we earn on qualifying purchases.

From Linux Modifications to UEFI

The tool follows earlier work on GPU compute unlocking. A prior project, GPU Unlock, developed by Zagitta, demonstrated nearly double the normal Tensor Core throughput on RTX 30 Series cards in certain FP16 and BF16 compute workloads, according to Wccftech’s earlier coverage. Reported AI image-generation benchmarks in that work showed improvements approaching 60%. That implementation relied on a Linux kernel module, and a hardware-specific experimental Windows version also emerged.

NV UEFI Unlocker takes a different route by applying its changes through UEFI before driver initialization and targeting multiple GPU generations and operating systems. Its temporary nature distinguishes it from permanent firmware flashing, but it does not remove the need to follow the project’s hardware-specific instructions. The reported benchmarks and compatibility details are based on early project information and testing, rather than a guarantee covering every configuration.

“The modification is temporary and disappears after a complete power-off.”

— NV UEFI Unlocker project documentation

Results Depend on Card and Workload

The project is still experimental, and successful unlocking is not assured on every GPU or system. Its compatibility listing identifies model-specific issues, and GTX 16 Series support remains experimental. RTX 50 Series cards are outside the stated supported range. The reported figures do not establish how consistently the tool performs across different operating systems, system configurations or applications.

It is also unclear how much of the synthetic throughput increase will carry over to particular AI tasks. Workload design, memory bandwidth and power constraints can affect real-world performance, and the reported benchmarks do not show equivalent results for every user. There is no convincing evidence in the available reporting that the changes improve gaming or DLSS execution times. Secure Boot requirements and the need for a bootable USB procedure may also make installation unsuitable for some users.

Further Compatibility Testing Needed

The next useful evidence will come from additional testing across supported GPU models, operating systems and real AI workloads. Users considering the utility should check the project’s current compatibility matrix and instructions before attempting the boot process, and should not assume that benchmark gains apply to their setup. Further reports may clarify which models work reliably and whether synthetic improvements translate into practical reductions in task time. No specific release date or next project milestone is established in the available reporting.

Source: Wccftech

Key Questions

Which NVIDIA GPUs are listed as supported?

The project lists GeForce RTX 20, 30 and 40 Series cards, although some models reportedly have issues. GTX 16 Series support is experimental, and RTX 50 Series GPUs are unsupported.

Does the utility permanently modify a GPU’s VBIOS?

No. The described approach runs through UEFI before the graphics driver initializes and does not require a permanent VBIOS flash or NVIDIA driver patch. The change is reported to disappear after a complete power-off.

How much faster are the reported benchmarks?

One reported RTX 2070 test rose from 36 TFLOPS to about 60–63 TFLOPS in FP32-accumulation matrix multiplication. An RTX 4060 Laptop GPU reportedly increased from 29.89 to 49.33 TFLOPS in FP16 compute. These are specific early benchmark results, not expected gains for every card or application.

Will it improve gaming or DLSS performance?

There is currently no convincing evidence that the modification improves gaming performance or DLSS execution times. The reported gains concern particular AI compute workloads.

What does installation require?

The documented process uses a bootable 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

You May Also Like

PCIe 5.0 Doubles PCIe 4.0’S Bandwidth, But Your Graphics Card Barely Cares

PCIe 5.0 doubles the theoretical bandwidth of PCIe 4.0, but tests show little gaming difference for most current graphics cards.

How Philips’ 36-Inch Curved Gaming Monitor Reaches Refresh Rates Of Up To 260 Hz

Philips has launched the Evnia 36M4C5500P, a 36-inch curved WQHD gaming monitor with a Fast VA panel and refresh rates up to 260 Hz.

Your 2026 Guide To 10 Graphics Cards For AI And Gaming

A 2026 roundup compares 10 graphics cards for gaming, AI workloads, compact PCs and basic display upgrades, with key fit and power tradeoffs.

NVMe SSDs Explained for Game Load Times

Learn what NVMe speeds really mean, why games rarely load twice as fast, and which SSD upgrade gives your gaming PC the biggest improvement.