GitHub user Zagitta has developed and released an open-source project called GPU Unlock that removes instruction issue rate limits on NVIDIA Ampere architecture GPUs through a Linux kernel module, unlocking hidden tensor core computing power. Real-world tests show performance gains of more than 60% on models such as the RTX 3090 Ti, and the project is experimental.