What's wrong with Intel? Core 2 was a great leap, i7 and Sandy Bridge offered massive performance and Ivy/Haswell drove down power consumption to much more manageable levels. Broadwell will finally bring it down to fanless levels on some ultra books.
Intel offer
- great process
- great circuits
- adequate microarchitecture because of
- lousy architecture
As a consequence Apple can match their performance at low frequency in smaller area, lower power, and less advanced process. Apple using 28nm and no FinFet can mostly match a Haswell i3 manufactured at 22nm.
Now this is not a fair comparison.
(a) Intel offer a few kickers to boost performance in some specialized cases, from hyperthreading and AVX2 to 4 (and more) cores to the ability to get to 4GHz.
(b) Apple could, if they wanted, easily enough throw in four cores (or five to match hyperthreading) and could extend their vector width. Those are minor tactical decisions based on the target market.
(c) More interesting is the frequency question. COULD Apple, if they wanted to ditch Intel (eg because of this issue of being linked to their update schedule) run their core at 4GHz (using, of course, desktop like 35 or 45 or 65W rather than phone-like 3W)? I honestly don't know. Doing this requires
- some microarchitectural tricks to split the pipeline nicely and to limit the damage from memory being so many cycles away (I'm confident Apple could do this)
- some circuitry tricks (not so sure)
- process tricks (not at all sure)
On Apple's side, they have a much simpler design target, and because of this they have the flexibility to try some daring ideas that may be too much for Intel (eg the CPU consists of two layers of silicon rather than one --- ie very simple "3D" design. Schemes like this allow multi-ported structures to split the ports between the two layers with consequent substantial (30% or so) improvements in speed and simultaneously lower power.)
There are also a number of tricks we know Intel use which Apple apparently aren't yet using, so they have a much clearer path ahead of them.
All indications are Apple get access to a 20nm process this year, and Finfets next year. Each of these should be good for around a 30% speed boost at the same power. Throw in some obvious micro-architectural improvements Apple is not yet using (eg split L2 caches and a much larger faster L3 cache, better memory controller) and I'd expect Apple to be able to generate at least 1.4x speed boost for the A8 and A9. Point is, Apple appears to be on its way to matching Intel's single-threaded performance a whole lot faster than Intel appears to be raising that performance.
Which is still VERY DIFFERENT from claiming that Apple can ship an Intel iMac replacement chip this year. But in late 2016???
