What's been impressive is that emulated code optimized for x86 runs faster on the M1. If more developer mindshare focuses on ARM optimizations and specifically writing software for the M1, the performance gains likely will even be more significant. That said, these chips are designed for a single computer, which hasn't been much of a focus area for Intel/AMD the past few years.
It'd be interesting to know how close the code quality is to what you'd get from generating ARM code from the compiler. The jump from compiling to the M1 may not be too big if the translation is good.
Another question is whether it's equally able to take advantage of the specialized execution units. It'd be interesting to know whether the native M1 compiler can target those units directly or if they're mostly used by Apple's libraries. I'd bet on the latter, although since there's no overhead of transferring data to the GPU, I wonder if there's any opportunity there to expand usage there.
The fact that little more than a recompile is needed to make native binaries (obviously lots of exceptions to that) is also impressive and a testament to the work done until now on iOS/iPhone/iPad ecosystem. Of course, some minor complaints that old 32-bit apps don't work, even on intel systems.
Meanwhile we're on year 8 of false starts for Windows on ARM. In contrast to apple, a 10 year old 32-bit x86 app might work, but a recent app probably won't.
It definitely isn't! The point was that we should consider both software + hardware together when praising the performance of the M1. That is why comparing the M1 with an Intel / AMD processor is like comparing Apples to Oranges. :)