As the smartphone market has moved to four, six, and even eight-core processors, Apple (NASDAQ:AAPL) has stuck with dual-core processors for the home-brewed A-series processors found inside its iPhones. Apple has remained focused on per-core performance (which benefits all applications) rather than on shoving as many cores as possible onto the chip in order to win multicore-capable performance tests.
I believe that even as Apple transitions to the much more power-efficient 14/16-nanometer FinFET manufacturing technologies for its iPhone-bound A9 chip, it will still stick with a dual-core processor for the processor.
Remember when Apple talked about "sustained performance"?
At the iPhone 6 launch earlier this year, Apple exec Phil Schiller spent an unusual amount of time talking about the A8 processor and, in particular, how it offers more sustained performance than the A7 did. Peak performance for the A8 did move up from the A7, but it wasn't the large gain we saw in moving from the A6 to the A7.
Ars Technica in September published an investigation of Apple's claims of improved sustained performance. Last year's A7 starts out running at 1.3 GHz, but after 10 minutes of heavy load the CPU throttles back to about 967 MHz. This year's A8, according to the article, starts at 1.4 GHz in the iPhone 6/6 Plus and then throttles down to between 1.167 GHz and 1.2 GHz.
This is a substantial improvement in sustained performance, and it tells us something about where Apple should go next with CPU cores in phones.
More frequency, better architecture, two cores
If Apple were to simply port over the current A8 processor to the 14/16-nanometer FinFET nodes, I'd imagine the company could get a significant boost in sustained performance just from higher clock frequencies at the same power. Add an enhanced chip design, and I think Apple could see a sustained performance boost of nearly 50%.
If Apple were to move to three or four cores in a phone, then in very short performance tests that use multiple cores it could brag about performance leadership. However, that kind of performance could not be sustained in the real world.
For most mobile apps, the two extra cores would just sit idle. For the rare mobile apps that can use more than two cores, frequencies would have to be dialed down in the cores in use. Total delivered performance might rise in this case, but it's entirely possible that allocating that additional power to two cores and making them run faster would ultimately be the better way forward.
Wait a minute ... Apple moved to three cores in the A8X!
Readers might recollect that Apple moved from two cores to three in the A8X that powers the iPad Air 2. This is true, but keep in mind that a larger tablet chassis can deal with far higher levels of heat than a small smartphone can. Look at this presentation slide (captured by the EE Times) from Taiwan Semiconductor (NYSE:TSM) for the International Electron Devices Meeting:
Notice that there is a pretty significant delta between the "smartphone sustained power envelope" and the "tablet sustained power envelope." This is likely why Apple felt it possible to move to three cores for the tablet, but stuck with two cores on the phone.
In moving to 16 FinFET+ on the A9, I believe Apple will stick with the same core configuration featured in the A8, but will improve the CPU design and move up the frequency curve, enabled by the new manufacturing technology.