With the Apple (NASDAQ:AAPL) A9 processor essentially a known quantity thanks to the various performance tests and tear-downs that have hit the Web over the last month or so, it's now time to look ahead to what Apple might do with its next-generation A10 processor, and in this article I'd like to put forth a couple of educated guesses about what I expect Apple to do with this chip.
Higher frequencies thanks to TSMC's 16-nanometer FinFET Plus
It is well known at this point that the A9 manufacturing contract is split between TSMC (NYSE:TSM) and Samsung (NASDAQOTH:SSNLF). This information was first reported by the China Times and has been verified in a number of other reports (I confirmed this information with a source I believe to be reliable, and famed KGI Securities analyst Ming-Chi Kuo independently reported the same).
At any rate, I am told that in terms of transistor performance, TSMC's 16-nanometer FinFET Plus is superior to Samsung's 14-nanometer LPP process (Apple is building the A9 on both).
Given that the TSMC process is the better-performing one, and given that Apple doesn't have to worry about making sure the A10 can be manufactured to specification on the Samsung process (i.e., a "least common denominator"), I think Apple will be able to realize a "freebie" increase in CPU/graphics core frequency as a result of building the A10 on a more mature, intrinsically higher-performing 16-nanometer FinFET Plus process.
Support for much greater memory bandwidth
The A9 chip's memory controller supports LPDDR4-3200 memory, which gives the system-on-chip an effective memory bandwidth of 25.6 gigabytes per second. Since I believe that Apple will want to significantly improve the graphics performance of the A10 over the A9 (more on that later, though), I think it will want try to realize a big boost in memory bandwidth (i.e., the rate at which data can be transferred between the system-on-chip and system memory).
Fortunately, this should be fairly straightforward for Apple to achieve. Samsung (NASDAQOTH:SSNLF) recently announced that it had begun mass production of LPDDR4-4266 speeds. This, as the memory giant notes in its press release, is a whopping 30% faster than the current LPDDR4-3200.
In a dual channel configuration, the A10 should be able to realize memory bandwidth of 33.28 gigabytes per second using this faster memory. This significantly increased memory bandwidth should allow Apple to scale up the performance of the rest of the SoC and worry less about bandwidth limitations.
More memory bandwidth means beefier graphics?
In AnandTech's review of the iPhone 6s, the reviewers noted that one of the reasons that Apple didn't move from a four-core graphics configuration to a six-core configuration until the A9 was due to the fact that the previous A-series chips were fairly bandwidth constrained.
If Apple goes ahead and uses this significantly faster memory as I expect it to, there could be room for Apple to upgrade the graphics configuration in the A10 from a six core PowerVR Series 7 design like the one in the A9 to an eight core Power VR Series 8 design.
Not only would Apple be able to deliver increased graphics performance by virtue of the architectural enhancements in a next-generation PowerVR GPU (or potentially even a custom Apple-designed GPU, since it is well-known that Apple has been working on one for quite a long time), but the additional cores would also help provide extra oomph.
This additional graphics performance will be especially important if Apple decides to increase the display resolutions on its next-generation iPhone 7/7 Plus phones, a move that is, quite frankly, a distinct possibility. Indeed, I suspect one of the reasons -- beyond additional power draw -- that Apple has been hesitant to move to even higher resolutions is that it wants to make sure that 3D games run well at the native resolutions of its devices.