This is a Linux kernel module that unlocks the FP16/BF16 instruction issue rate of NVIDIA RTX 30 series consumer GPUs to the level of their professional counterparts, significantly boosting matrix operation throughput. For example, the BF16 compute performance of the RTX 3090 Ti can increase from 87.5 to 157.4 TFLOPS, and large model prefill speed improves by 14%–36%. It performs the unlock before the official NVIDIA driver loads, requiring no driver modifications and no need to reapply patches after each driver update.