What are the key differences between CPUs now are CPUs twenty years ago? Sure we can pack more transistors into a smaller space, but clock speed has plateaued. Are all the innovation purely in pipelines and branch prediction?
In the spirit of increasingly large memory caches, don't forget other increasingly large microarchitectural buffers (e.g. reorder/retire buffer), and the hella microarchitectural parallelism that it enables. Apple went all in on this strategy, and it paid off.
I think clever choice of instruction set is another one. Today's x86-64 is a far cry from what an 8088 could do, and Apple's Silicon does more in one clock cycle than its predecessors by a mile.
