▲ 1k today3.1k starsCHN ×1New · 3d old
Linux kernel module doubles RTX 4090 tensor throughput
What is gpu-unlock-ada?
The GPU-unlock driver patches the Nvidia SM speed-select before the official driver loads, letting the RTX 4090 issue FP16 and BF16 tensor ops at roughly twice the stock rate. It also supports RTX 4070 Ti SUPER and RTX 4060 Ti 16 GB on Linux.
Who it is for
ML researchers and hobbyists running CUDA on Linux can squeeze far more performance from existing cards without any hardware changes.