GPU-unlock: FP/BF16 instruction issue rate on Nvidia 3000's cards

A Linux kernel module called GPU-unlock doubles the FP/BF16 instruction issue rate on Nvidia 3000-series graphics cards (GA102, GA103, GA104, GA106) to match their professional equivalents by running before the NVIDIA driver loads. The unlock significantly improves performance—for example, boosting BF16 performance on a 3090 Ti from 87.50 to 157.41 TFLOPS—without requiring manual patching with each new driver release. Installation involves cloning the repository, running make install, and adding a kernel command line parameter.