Hey guys. I'm new. I think it is more interesting to talk about the technology behind Intel than to watch it's boring share price move
Anyway, regarding the topic A7 vs. Silvermont, I think that's quite interesting since those are basically using the opposite approach to stay in the same power envelope while maximizing compute power.
First of all, it's 2 vs. 4 cores. I very much prefer the approach Apple chose: Increase the single thread performance and use two cores instead of 4 slower cores. That makes so much sense with IOS and it would also make a lot of sense for Android. The thing is, these OSes have been designed from the start to reduce multitasking to a minimum. Apps are put into some kind of hibernation while in background, something introduced by IOS and copied by Android. Therefore, in most scenarios, at least two of the four cores of a quadcore CPU are not used at all. Usually it's the OS which always runs in the background together with the App which is currently in the foreground (certainly, both rarely use 100% of one CPU all the time). If you consider, that even with full blown PC OSes like Windows, 4 cores are rarely busy at the same time, you could think of how often this would be the case with non multitasking mobile OSes like IOS and Android. To me, it is simply a waste of die area and power to build quad cores into smartphones (and also tablets that use these types of OSes). The problem is that most people, just like it's the case with 64 bit, think it's better. It's a marketing thing which Apple doesn't care about. They build what's best and most of their costumers trust them in that.
Sure, some people may argue that it's possible to use multiple cores via multithreading by a single process, but we all know that this is rarely the case. Even browsing the web today, which is quite processing extensive, multicore doesn't help much in terms of speed just because parallelization of software is difficult and costs much effort that often isn't spend by programmers - something that hardware simply can't fix.
Anyway, so Apple's design decision is better than Intels, though I absolutely understand that Intel went this way since they have to compete with feature lists of other phones where the high end devices claim quadcore processing power. Actually I think the initial approach Intel took with Medfield, single core with hyperthreading, actually would be the best in terms of real world benefit. Making this single core as fast as possible would, in my opinion, deliver the best user experience by far of most real world application. The problem is that people just don't want to buy single cores.
So, continueing from here to the question, what's better, 2 issue (Silvermont) or 6 issue (rumours (!) say that's A7). I think that's a difficult and multidimensional question. The 6 issue design of the A7 allows to deliver high performance when automatic parallelization of the code works good (which is not always the case). What the A7 gains is the possibility to use a low clock and therefore the possibility to use a lower voltage, reducing power consumption significantly (it's to the power of 2). What they also gain, and that is equally important, is the possibility to use a shorter pipeline. That helps especially with branch misses, which either lower performance (e.g. in loops, which are very common in basically all code) or are prevented by a branch prediction, which itself is often a very complex and therefore also power consuming circuit. On the other hand, 6 execution units often execute code that isn't actually needed and that's a waste of power but improves throughput (a lot, depending on the code structure).
Intel went the other way, already with the first Atom core. They used an in order (single) execution unit which makes for a simple design and reduces complexity/power consumption. They used a high clock and introduced hyperthreading to compensate on the lower processing power. But, as always, it has a drawback: For an increase of the processing speed, they had to increase the clock frequency and therefore increase the voltage and increase the pipeline length.
So what can be done with these approaches to lower the power consumption? Apple could use some smart circuits, e.g. a smaller prediction circuit that tries to identify situations where some of the 6 execution units could be clock gated so they don't consume additional power when actually not providing a benefit for the currently processed code snippet. There are many other ways to improve the power consumption in such a core but none are trivial for sure.
Intel's approach, on the other hand, basically needs improvements on the semiconducter process to lower dynamic and static power consumption and allow for reduction of the pipeline depth (that's basically a matter of transistor switching speed). There are certainly things that can and have to be done on the circuit design level but their approach seems to be relying much more on a process advantage, which makes much sense for Intel.
What I actually wanted to say is that this is not a black or white or good or bad decision. It's a very complex and multidimensional one which is not easy to answer in a simple way of a better or worse general approach.
I am working in the hearing aid industry and we are able to run a DSP with 0.05 mA when idle (including SRAMs) and 0.2 mA when running full load (including SRAMs). It is clear that this is a low clocked and simple designed DSP with low gate count and a short pipeline. It's doesn't deliver the highest compute performance per clock but it delivers a very good performance per watt. If you want to improve the former, you have to sacrifice the latter. Simplicity wins here.
I bet that none of these modern cores, be it Silvermont, A7 or Haswell, beats an ARM M0 in terms of performance/watt even closely.