Windows 11 users can unlock more AI performance on NVIDIA RTX 20, 30, and 40 GPUs

The open-source nv-uefi-unlocker tool supports NVIDIA GeForce RTX 20, 30, and 40 series GPUs on Windows and Linux.
The developer reports up to 2.6x higher Tensor Core throughput and up to 94 percent better performance in certain DirectML inference operators.
The UEFI-based tool runs from a bootable USB drive without flashing the graphics card’s VBIOS, but you must repeat the unlock after every cold boot.
The tool targets AI workloads, not gaming, and comes with GPU compatibility, boot, power, and thermal considerations.

Windows 11 users with older NVIDIA GeForce graphics cards could get more AI computing performance from their existing hardware thanks to a new open-source tool that unlocks additional Tensor Core performance.

The project, called “nv-uefi-unlocker,” combines several previously developed NVIDIA GPU-unlocking techniques into one automated, UEFI-based tool. It supports GeForce RTX 20, RTX 30, and RTX 40 series graphics cards, including desktop and laptop models.

The goal of this project is to help users get more performance from AI workloads running locally on their computers, including image and video generation, DirectML inference, and large language models, without buying a newer graphics card.

According to the developer’s measurements, the modifications can increase FP16, BF16, and TF32 Tensor Core throughput by up to approximately 2.6 times. FP8 throughput can improve by up to approximately 1.9 times.

The project also reports 14 to 75 percent improvements in image and video generation workloads, up to 94 percent in DirectML inference operators, and up to 19 percent in large language model inference prefill performance. The developer reports a smaller improvement of approximately 5 percent for LLM decoding.

These figures represent different workloads and testing conditions, not a universal performance improvement. Actual results depend on the graphics card, application, precision format, memory bandwidth, and power limits. The LLM measurements, in particular, were obtained using a dual-GPU mixture-of-experts configuration and should not be interpreted as representative of every local AI setup.

More AI performance without upgrading your NVIDIA GPU

The main appeal of nv-uefi-unlocker is that Windows 11 users can get more AI computing performance from graphics cards they already own instead of upgrading to newer hardware.

NVIDIA Tensor Cores are specialized processing units designed to accelerate calculations used in AI and Machine Learning (ML). These operations are important for apps that generate images and videos, run AI models, and run large language models locally.

The project targets certain performance restrictions affecting Tensor Core computing. According to its developer, supported graphics cards can deliver significantly higher throughput for specific precision formats after using the tool.

For example, the developer reports that FP16, BF16, and TF32 performance can reach approximately 2.6 times the original throughput in some tests. However, power consumption, memory bandwidth, and the graphics card’s hardware configuration generally limit real-world gains.

The reported improvements are particularly interesting for users who rely on local AI applications rather than cloud-based services. However, apps must use the affected GPU computing paths to benefit from the changes.

It’s also important to understand that this is not a gaming performance upgrade. The tool targets AI computing workloads and does not promise higher frame rates, better gaming performance, or faster general graphics processing.

Which NVIDIA graphics cards are supported?

The tool currently targets three major GeForce generations.

RTX 20 series (Turing): Supported, although some models have known compatibility issues. Windows users must also complete an additional GSP firmware configuration step before running the tool.
RTX 30 series (Ampere): Supported across desktop and laptop graphics cards based on the GA10x architecture.
RTX 40 series (Ada): Supported across desktop and laptop graphics cards.

The project also lists GTX 16 series graphics cards as theoretically compatible, but it needs additional real-world testing.

NVIDIA’s RTX 50 series, Pascal and older architectures, and GA100-based graphics cards are not supported.

Compatibility at the GPU-family level does not mean every individual graphics card has been independently tested. The developer maintains a list of supported graphics cards, including model-specific test results and known limitations.

How the NVIDIA AI performance unlocker works

Unlike a conventional Windows application, nv-uefi-unlocker works before the operating system loads the NVIDIA graphics driver.

The tool uses a UEFI-based process that applies the required changes during system startup. Instead of permanently modifying the graphics card’s VBIOS, it uses a bootable USB flash drive to unlock it.

Users first download the project distribution package and prepare a FAT32-formatted USB drive. The included scripts identify the graphics card generation, download the required NVIDIA firmware files, verify them, and generate the necessary boot files.

The computer must support UEFI booting, and Secure Boot must be disabled unless the user configures a trusted self-signed certificate.

After preparing the bootable media, users must restart the computer and select the USB device from the firmware’s one-time boot menu. The tool then applies the changes and automatically hands control back to the operating system.

The project supports Windows and Linux, though some prerequisites differ by GPU generation and operating system.

For example, RTX 20 series Windows users must first run the included PowerShell script to configure the required GSP firmware. Linux users have additional GSP-RM and power-management requirements.

One important limitation is that you must repeat the unlocking process after every cold boot. The changes do not permanently modify the graphics card’s VBIOS, and the unlock state is not retained after power loss.

The developer provides diagnostic utilities to compare performance before and after the modification under the same workload.

This is not a gaming upgrade, and there are risks

The tool focuses on Tensor Core computing for AI workloads, not gaming. The project does not claim improvements in gaming performance or frame rates.

The developer also notes potential black screens, frozen displays, and boot errors on certain configurations. Additional power consumption and heat are concerns, particularly when combining the unlocker with overclocking or power-limit modifications. Users should check the compatibility matrix and follow the recovery instructions before attempting the modification.

Where to download nv-uefi-unlocker

The project is available on GitHub. The package does not include the NVIDIA firmware. Instead, the included scripts download the required files from NVIDIA’s public release packages and verify them locally.

Would you unlock additional AI performance on your existing NVIDIA graphics card?

Yes, I'd like more local AI performance.

Maybe, if my GPU is compatible.

No, the firmware risks aren't worth it.

I would rather upgrade my GPU.

Voting closes: November 9, 2026 1:00 pm

The repository also includes installation instructions, performance measurements, diagnostic utilities, a list of supported graphics cards, and recovery guidance.

For Windows 11 users who want to run AI applications locally, nv-uefi-unlocker offers an interesting way to explore extra performance from older NVIDIA GeForce graphics cards without flashing the VBIOS or buying newer hardware.

However, the reported improvements depend on the workload. The modification doesn’t improve gaming performance, and the firmware-level process carries compatibility and stability risks. It’s best suited to enthusiasts who understand the requirements and are comfortable experimenting with their hardware.