Because they have moved past the easy parts of optimization. Apple's optimizations still have thermal throttling issues (sustained use of an A11 will not beat an Intel chip due to better cooling and IPC).
Compared to other ARM licencees they do beat the pants off of. This is probably why Apple put in a T1 chip and a touch bar in the Macbook Pro and the T2 chip (dedicated power module) in the iMac Pro. They are prototyping the individual pieces with Intel as the main processor so that they can attempt to swap the chip in 2020.
sustained use of an A11 will not beat an Intel chip due to better cooling and IPC
In an iPad profile sure the heat dissipation is very limited. With even a basic passive heat sink (not just a spreader, but actual fins for surface area), much less active cooling like most laptop and desktops use, and the heat profile dramatically changes.
This change was inevitable -- recent A## chips have been just astonishing for passively cooled chips. The state of the art in binary translation, and the support for cross compiling across architectures, makes this a very doable transition.
That's a good joke and I laughed, but this seriously understates the power of Apple's ARM chips. They are pretty impressive. To put this on the pizza scale, Apple is a lot further along than any individual pizza maker. They're a strong regional pizza chain, at the very least.
Intel painted itself into a technological corner, where they have to pursue yields with larger die sizes than AMD. Or, looking at it from the other side, AMD won by steering themselves a course where they could avoid the trap of ever larger die sizes.
Single threaded CPU performance improvements are pretty much tapped out at this point. Process improvements have slowed to a glacial crawl (compared to the past), clockspeed can't really increase much while maintaining the ability to be air cooled, and IPC is mostly mined out. Intel probably has the best processor cache in the industry, and adding more just isn't yielding much for general purpose applications (can't really speak about multi-tenant environments or other specialized applications).
I'd guess, but I don't know, that Apple's performance growth is coming from two places:
1. More multi-threading. Moble chips in general have been getting more parallel over time. It's pretty easy to construct synthetic benchmarks that benefit from this; it's less easy for general purpose applications, although they are coming along.
2. The GPU is improving. It seems like there is a lot of headroom here to improve. Apple's GPU is almost certainly substantially better than Intel's. This isn't really a huge problem for Intel, because for people who care, they can always get a discrete GPU. Super powerful GPUs can be very impactful for certain applications, but they just don't have general applicability.
In general, I'm pretty skeptical that Apple's CPUs are improving in the necessary areas to enable them to switch their pro lines. Moving over their Air and Macbook products to in house chips seems pretty straight forward.
Because x86 was technologically mature in the 90s (superscalar, OoO, vector extensions, …) and basically hit its frequency ceiling in the mid-aughts (modern Core-series finally have frequencies beyond what P4 reached back in 2005~2006 but it took a long time).
ARM was nowhere near there (for it had no need to) before modern smartphones. ARM in 2007 was single-issue in-order single-core topping out at ~600MHz.
Standard math failure. It's the same with programming languages. I never trust anything that got a "2x" speedup because if a gaping optimization like that exists, it's likely not ready for prime time.
Anyone want to chime in on why Intel can't get more than 5% YoY while Apple has been getting 30-40%?