nv-uefi-unlocker Unlocks GeForce Tensor Throughput

This is a striking low-level performance unlock, but “2× AI compute” is a best-case benchmark result—not a universal speedup for every model or application.

• –Supports Windows, Linux, and other UEFI-bootable systems across Turing, Ampere, and Ada GPUs.
• –The unlock is volatile: it clears after power loss and requires an experimental pre-OS boot workflow.
• –Independent RTX 4090 testing found nearly 2× TF32 gains, but only 1.1–1.3× for FP8/FP16 and roughly flat BF16 results. [GitHub issue](https://github.com/physicsdolphin/nv-uefi-unlocker/issues/80)
• –Real workloads such as image generation can improve, though driver kernel selection and power limits remain major bottlenecks. [XenoSpectrum](https://xenospectrum.com/en/nvidia-uefi-windows-tensor-unlock/)
• –The biggest value is extending the useful life of existing GPUs for local inference and AI development, with firmware compatibility and stability still the main risks.