NV-UEFI-Unlocker Tool Boosts Nvidia GPU Tensor Performance

A new open-source UEFI tool boots from USB and unlocks significant extra tensor performance on RTX 20, 30, and 40-series desktop and laptop GPUs, making the half-speed AI throttle hack a whole lot easier to try on Windows ^[1]. The nv-uefi-unlocker project removes Nvidia’s tensor limiter using the community V67 payload with read-back verification ^[1].

Key facts * The nv-uefi-unlocker project covers RTX 20, 30, and 40-series desktop and laptop GPUs ^[1]. * The tool runs from a FAT32 stick before the Nvidia driver loads at every boot, requiring no VBIOS flash or patched driver ^[1]. * The developer’s own laptop RTX 4060 hit 65% higher FP16 math using this method ^[1]. * The tool allows the unlock to run entirely from a bootable USB drive on both Windows and Linux ^[1].

How does the nv-uefi-unlocker work? The nv-uefi-unlocker project tricks Nvidia’s signed Booter firmware into opening the write lock on two speed-selector registers, allowing the tensor units to run at full rate ^[1]. This process uses a community V67 payload with read-back verification to ensure the unlock is safer than previous experimental builds ^[1].

What are the limitations and requirements? Results from the unlock are contingent on power and thermal limits, and the project warns that combining it with the nvoc power-limit unlock "can exceed factory power and thermal envelopes" ^[1]. The project also notes that a per-card ROM hash check is not yet built, and the unlock’s persistence on Thunderbolt eGPUs is not guaranteed ^[1]. Additionally, the original creator of the gpu-unlock project, Zagitta, does not plan to maintain the Windows version and warns users to be careful with unofficial installers ^[1].