NV UEFI Unlocker Boosts AI Performance by Up to 75% on RTX 20, 30, and 40 GPUs - eTeknix

The tool works by modifying how the GPU functions before the NVIDIA driver loads in the operating system, using a UEFI booting method. Unlike traditional firmware modifications, this solution doesn't require permanently flashing the VBIOS or patching drivers, making it completely temporary and reversible when the PC is shut down.

Building on Previous Success

NV UEFI Unlocker builds on an earlier project called GPU Unlock by Zagitta, which had already shown improvements of nearly 100% in Tensor Core performance and up to 60% in AI image generation. However, that project was primarily limited to Linux systems. With this new UEFI method, these benefits are now accessible to Windows users.

Performance Gains

Testing has shown significant performance increases. An RTX 2070 achieved between 60 and 63 TFLOPS in FP32 matrix multiplication operations, up from 36 TFLOPS—a 67-75% improvement. A laptop RTX 4060 demonstrated a 65% increase in FP16 performance, rising from 29.89 to 49.33 TFLOPS.

Implementation Considerations

Users should note that the process requires preparing a bootable USB drive, disabling Secure Boot (or configuring its keys), and enabling GSP firmware mode for RTX 20 series cards. Additionally, while the tool boosts AI computing power, it doesn't improve gaming or DLSS performance, as actual gains depend on memory bandwidth and workload type.

Although still experimental, NV UEFI Unlocker provides a way to extract more performance from existing hardware for local AI tasks without additional investment, making it particularly valuable for enthusiasts and professionals working with AI applications.