For the vast majority of software this is literally just a build option. For things which really need specific chip instruction sets, then that's where you get into interesting territory - a simple rebuild will yield poor performance.. probably.
Absolutely. If software has, either in itself or in any of its dependencies, gone and made heavy use of AVX SIMD extensions, then ARM is going to be something of a disappointment. AFAIK there's no ARM that has a vector unit anywhere near as grunty as Intel's AVX.
This was a problem last time Apple switched architectures. Software like Photoshop had been heavily optimised on PowerPC, making very effective use of that architecture's Altivec SIMD vector processor. At the time, Intel were all over the place (MMX, SSE, lord knows what else), and Photoshop for Intel was substantially slower at some tasks. It took quite some time for Intel to finally catch up with Altivec.
[Intel didn't fully match Altivec until around about 2012, which is when I think that Intel finally added an FMA instruction to their x64 line up. Prior to then they'd had an FMA instruction on Itanium processors, and the suspicion was that they weren't putting one into x64 in the desparation to give Itanium an artificial edge. By 2012 they'd given up with that, for they'd finally read the writing on the wall which customers had carved in letters 10 feet high and 4 inches deep.]
Another aspect is that there's way more to it than simply shoving in a comparable sized SIMD unit alongside an ARM core. To keep such a processing monster fed you need a very extensive cache system. So you end up with a chip that's pretty much as monstrous as Intel's are (in terms of transistor numbers). The savings come from not having all the transistors dedicated to pipelining x86 opcodes which will save silicon space / current / heat. Also, Apple can get ARMs made on TSMC's 5nm process. In contrast there's no sign yet that Intel have mastered their 10nm process (equiv to TSMC's 7nm), let alone worked out how to get smaller still.
Apple could have got x64 processors from AMD, which will get made on TSMC's 5nm process; in fact, they could buy AMD with some loose change. If they're not doing that, and are going all-in on their own ARM; well I guess they've got the money to do that, and it would allow them to forge their own path from here on.
Biting the hand that feeds IT
