There is an fundamental complexity in the semantics of x86_64 instructions that makes them expensive to implement. I worked for Transmeta and am at another mic

There is an fundamental complexity in the semantics of x86_64 instructions that makes them expensive to implement. I worked for Transmeta and am at another microprocessor company today so I do have 1st hand experience. There are other things in M1 that helps make M1-based systems fast. I'm not talking about that. I'm exclusively referring to pure-core performance. It is true that the latency improvement and high-bandwidth of the nearby DRAM does help here, but it is far from the major factor. FWIW, there's no evidence that TSMC's N5 node is the major factor, beyond letting Apple spend a lot of transistors. Intel/AMD have so far been able to overcome the overhead with lots transistors and power, afforded by leading edge nodes. Apple is now at the same playing field and especially the power efficiency difference is obvious.