So, I was thinking about what the future holds for deep learning hardware. It seems like nvidia dominates, but big challengers like Intel (and Nervana!), AMD , Xilinx, etc. are coming up with their own products. Startups like wave computing and graphcore also appear to be doing interesting stuff. I'm very interested in FPGAs for deep learning and even more about ASICs and special chips for deep learning. Are there any papers and/or companies you could point me to? What do you guys think about the future of deep learning hardware? I understand that GPUs are already very good because of matrix multipliers and FPUs, but surely an opportunity exists just by lowering precision (for inference mainly, but apparently stochastic rounding works for training also)?