Comment by mmm_grayons

Can anyone who knows about machine learning hardware comment on how much faster dedicated hardware is as opposed to, say, a vulkan compute shader?