IPhone 7 uses A10 Fusion microprocessor chip. Specifically, they have four cores, seamlessly integrating two high-performance cores, and two high-efficiency cores that are capable of running at just one-fifth the power of the high-performance cores. The microprocessor is TSMC 16FF-based and specifically a APL1W24 processor. It runs around 2.33 GHz which is roughly 25 percent faster than the 1.85 GHz of the previous architecture.
A piece of standalone silicon is introduced in its new AirPods, dubbed the W1 chip. In total, this represents a great deal of engineering work done by Apple over the last year, and the A10 is the most significant to Apple’s system-on-a-chip (SoC) line since the company’s transition to 64-bit.
Apple unveiled the biggest technical changes featured in the A10 at the very beginning, boasting a four-core CPU with 3.3 billion transistors. The transistor is very likely to fell somewhere in the middle between the 2 billion count on the A8 and the 3.3 billion of the new A10. A transistor count well under 3 billion seems probable for the previous architecture. The 3.3 billion number for the A10 is well over 50 percent larger than the previous architecture, and the large jump is likely mostly thanks to the addition of two new, albeit small, CPU cores along with a greatly enhanced image signal processor (ISP). The GPU remains a six-cluster design.
The process node is not different from the A9 fabricated on TSMC’s 16nm FinFET process. The logic board means a larger device package. A10 fusion peak performance is up to 40 percent greater than the previous architecture.
A 25 percent clock speed increase is significant given that the increase was likely enabled by the better thermal performance of InFO packaging. It is also likely only possible because of Apple’s heterogeneous architecture which now features a pair of high-speed cores along with a pair of low-speed, power conscious cores.
Apple’s clock speed increase is probably more than just turning up the dial on voltage to make the cores run faster. By introducing the pair of low-speed cores, Apple opened up a whole new spectrum of dynamic voltage and frequency scaling (DVFS) options for completely disabling cores or their sub-parts. Apple designed its own performance controller to manage workloads between the cores, and we know from some industry sources that Apple does cache-sharing so that the caches don’t have to constantly read each others’ contents to be ready for a switch lest they face a delay in getting current data when they switch on.
The boost to 2.33 GHz clock speeds brings Apple much closer to the clock speeds of competitors from SoC makers such as Qualcomm and Samsung, and Apple may also have made some transistor changes to reach those speeds. By increasing voltage, and choosing transistors with higher static leakage (unavoidable waste power), Apple can get to these higher clock speeds. Apple’s chip team can also make architectural designs that have a higher power footprint in general, whether it be at a higher transistor count, more management power overhead, or more switching activity through a different logic implementation.
Apple’s two small cores in the A10 have drawn just as much interest as their larger cousins, with a lot of speculation centered on whether they too are an Apple custom design, or if they are a variant of a stock low-power core from ARM, such as the Cortex-A53.
Apple’s main core designs had been optimized so much that there were few gains to be had, and those gains would have been with serious diminishing returns. Ratcheting up the clock speed is an easy way to get more performance, but the thermal and power costs associated with that may have been the driving force for the split. The graphics power of the A10 is such that the GPU can be up to 50 percent faster than the previous architecture GPU while consuming only 2/3 the power. Since the announcement of the 7XT series of GPUs from Imagination Technologies that was featured in the previous architecture which was simply to add computer vision and compute performance enhancements to the existing 7XT line.
The power reduction alone rules out that Apple increased clock speeds to make these performance claims, so we are likely looking at some significant changes that feature an unannounced GPU, an Apple-designed GPU, or some other major architectural shift that we don’t know about. It is possible Apple could claim some gains through enhancements in metal, but up to 50 percent improvement in speeds seems a rather high claim for that.
Apple’s performance boost claims have historically tended to actually show in benchmarks, so this will be an area of particular interest when the GPU gets fully benchmarked and pictured under a microscope.
The introduction of Apple’s AirPods was also an important moment because they feature Apple’s new W1 wireless connectivity chip. This is a very tough sector to get into and be a competitor in the general marketplace, however, as seen with Intel’s own LTE offerings likely featured in the new iPhone, for example. Rather than being built from the ground up, those chips are the product of Intel’s acquisition of Infineon and manufactured on a TSMC process rather than Intel’s own. The A10 chip inside the iPhone 7 comfortably outpaces its predecessors and Android rivals, and even outdoes a wide catalog of relatively recent Mac computers.