Open-source gpu-unlock lifts Tensor Core rate cap on RTX 30-series GPUs

Zagitta/gpu-unlock is an open-source Linux kernel module that removes Nvidia's firmware cap on FP16/BF16/TF32 Tensor Core throughput for RTX 30-series cards during initramfs, without VBIOS flashing or driver patching. On an RTX 3090 Ti, BF16 GEMM rises from ~87.5 to ~157.4 TFLOPS (+79.9%) and TF32 from ~42.6 to ~82.6 TFLOPS (+94%).

• BF16 GEMM on RTX 3090 Ti: ~87.5 → ~157.4 TFLOPS (+79.9%)
• TF32 GEMM: ~42.6 → ~82.6 TFLOPS (+94%)
• FP16 with FP32 accumulate: ~87.5 → ~118.4 TFLOPS (+35.4%)
• Covers GA102, GA103, GA104 and GA106; installs via DKMS