Windows UEFI Tool Removes Tensor Compute Limit on Some GeForce RTX Cards

Technology · First reported 30 Oct, 15:59 · Updated 16:25

Image: Wikimedia Commons · NASA Johnson Space Center · Public domain

A Windows UEFI utility aims to lift NVIDIA’s tensor-compute limit on GeForce RTX 20, 30 and 40 series cards. One reported RTX 4070 Ti Super test showed higher FP16 throughput, alongside increased power use; the tool is aimed at local AI workloads, not games.

The tool is for Windows and targets GeForce RTX 20, 30 and 40 series cards; RTX 50 is not supported.
A reported RTX 4070 Ti Super test rose from 57.00 to 150.93 TFLOPS in FP16, while board power increased from 227 W to nearly 292 W.
The unlock resets at power-off and requires a USB stick on each cold boot; the report says games and regular FP32 CUDA work see no change.

Summary written by Lux AI from 2 articles

Key facts

Supported generations: GeForce RTX 20, 30 and 40 series; RTX 50 is unsupported.
Reported RTX 4070 Ti Super result: FP16 throughput rose from 57.00 to 150.93 TFLOPS; board power rose from 227 W to nearly 292 W.
Cold-boot requirement: Unlock resets at power-off and runs from a USB stick on each cold boot.
Intended use: Local AI tasks such as image generation and fine-tuning; games and regular FP32 CUDA work see no change.