I should have been more clear. I didn't mean the hardware, but the speedup you get from using mixed precision in something like Tensorflow with an NVIDIA GPU.