To be fair, I found another earlier ML preview to be more balanced (please ignore the title): https://towardsdatascience.com/apples-new-m1-chip-is-a-machine-lea

To be fair, I found another earlier ML preview to be more balanced (please ignore the title): https://towardsdatascience.com/apples-new-m1-chip-is-a-machine-learning-beast-70ca8bfa6203 https://towardsdatascience.com/apples-new-m1-chip-is-a-machi... As evident from the benchmarks, the result is very dependant on your network. Some getting huge acceleration boost from the neural processor, while others can't be accelerated. I would suggest try and see approach.