USB tool lifts Nvidia's half-speed tensor cap on RTX 20, 30 and 40 cards in Windows

Here is what is physically different: nothing in the card changes. A small UEFI program, run from a USB stick before Windows starts, tells the GPU to stop holding back its tensor cores, the units that do the matrix math behind AI work. The tool is called nv-uefi-unlocker, and it comes from Chinese modder physicsdolphin.¹²

The limiter works like this. On GeForce cards, Nvidia runs FP16, BF16 and TF32 tensor math with FP32 accumulation (the way partial results are added up) at half rate. The same silicon in professional cards runs at full speed.² The unlocker removes that cap before the graphics driver loads, and it doesn’t flash the VBIOS (the card’s own firmware) or patch the driver.²³

The idea isn’t new. A Linux-only version from Zagitta showed roughly double the tensor throughput on RTX 30 cards, with real-world AI image generation improving by nearly 60%.²³ An earlier Windows port from impacifist targeted a single RTX 3090 system. This tool covers RTX 20, 30 and 40-series cards on Windows, Linux and other UEFI-bootable operating systems.²³

The reported numbers vary by card:

• An RTX 4070 Ti Super went from 57.00 to 150.93 TFLOPS in FP16, a 2.65x jump, while board power rose from 227 W to nearly 292 W.²
• An RTX 2070 went from 36 to roughly 60 to 63 TFLOPS, a 67 to 75% gain.²³
• An RTX 4060 Laptop GPU climbed from 29.89 to 49.33 TFLOPS in FP16, a 65% gain.³

Those are user reports, not lab tests, and I’d ask what the setup was before trusting any of them. The spread is wide, from 65% to 2.65x, and that’s exactly why. Higher theoretical throughput also depends on the workload, memory bandwidth and power limits to show up in real use.³

There are strings attached. The unlock resets at every power-off, so the USB stick has to boot on each cold start. Secure Boot must be disabled or set to trust the tool’s certificate, which could bite games whose anti-cheat wants Secure Boot.²³ RTX 20 owners must turn on GSP firmware mode first. RTX 50-series isn’t supported, and GTX 16-series support is experimental.²³ Not every card has been tested.²

Gamers shouldn’t expect anything. Games and ordinary FP32 CUDA work see no change, and there’s no convincing evidence of better gaming or DLSS performance.²³ This is aimed at local AI work such as image generation and fine-tuning. The developer also warns that combining it with a power-limit unlock can push power and heat well beyond stock limits.²

Our read: The interesting part is how little hardware is involved. A tweak applied at boot, with no permanent change to the card, doubles a headline spec on some cards. For everyone else, it’s a curiosity. Watch the power draw: 65 extra watts on one card is real heat.

On the bench: if you run local AI models on an RTX 20, 30 or 40 card, this could cut your wait times, but check temperatures, expect to boot the USB stick every cold start, and plan around Secure Boot.

Footnotes

1. Tom’s Hardware, “Open-source USB tool unlocks Nvidia’s half-speed GPU AI limiter on Windows — lifts GPU’s FP16 math performance up to 65%, UEFI unlocker covers RTX 20, 30, and 40-series desktops and laptops”. Retrieved November 1, 2026. ↩

2. TechPowerUp, “New UEFI Tool Lifts NVIDIA’s Tensor Cap on RTX 20 to 40 Series Cards in Windows”. Retrieved November 1, 2026. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹²

3. Wccftech, “New NVIDIA UEFI GPU Unlocker Brings Up To 75% Higher AI Compute Performance To RTX 20, 30 & 40 Series GPUs On Windows”. Retrieved November 1, 2026. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹