A 3 layer convnet probably isnt the last word in whether the M1 is better for "deep learning". Something nontrivial where the gpu/processor is maxed out would b

A 3 layer convnet probably isnt the last word in whether the M1 is better for "deep learning". Something nontrivial where the gpu/processor is maxed out would be a more interesting comparison, even one of the pytorch vision tutorial that uses resnet I think could be a more realistic DL comparison