I just discovered the project https://github.com/ggerganov/whisper.cpp

As written in another comment, there is no way for (recent) CPU (even ARM ones) to be as fast as (recent) GPU on such big model (the list no GPU support in limitations).

https://www.reddit.com/r/MachineLearning/comments/10xp54e/comment/j7tk4fx/?utm_source=share&utm_medium=web2x&context=3

That being said, the project looks super cool, tks for the pointer (I ordered a M2 Max, lots of fun to come :-) )