Do anyone know how the two NPU compares to each other? The A18 Pro NPU is supposed to be twice as fast a the one in A17 Pro. How does it compare to the one in M4, is it still faster?
The first Geekbench results for the A18 Pro have it roughly at the same performance as the base M1, although that does not touch the NPU at all. NPU performance for A18 Pro could be around the same level as M2 or M3, given that M1 is the oldest supported SoC for Apple Intelligence on Mac and iPad and A17 Pro supports AI as well.
In GB6 ML tests the A18 seems to be 20-25% faster compared to A17 Pro. I wonder whether this is mostly memory bandwidth improvements or whether the NPU saw some refinement too.