Apple’s A8 SoC analyzed: The iPhone 6 chip is a 2-billion-transistor 20nm monster

2 billion transistors

Powering both the new iPhone 6 and iPhone 6 Plus is Apple’s new A8 SoC. This new chip, which contains around two billion transistors, will be one of the first mass-market components to use TSMC’s 20nm process. The A8 SoC will, according to Apple, be about 25% faster than its predecessor in CPU tasks, and 50% faster on the GPU side of things. These are some big performance gains that are made even more impressive by the fact that the A8 is reportedly 50% more power efficient than the A7 — and, despite almost doubling the transistor count, the die size is somehow 13% smaller. Read on for some detailed analysis of Apple’s new A8 SoC.

While Apple shied away from giving too many technical details about the A8 SoC, it did note that there’s around 2 billion transistors in there, fabricated with a 20nm process. This is a huge step up from the A7 SoC, which had “more than 1 billion transistors” on a 28nm process. While it’s possible that the A7 actually had significantly more than 1 billion transistors, and Apple was just hiding the fact, it seems unlikely. In either case, 2 billion transistors is a huge number for a mobile chip. Intel’s desktop Haswell chips actually have less transistors (~1.4 billion) than the A8 — though with a fully decked out GT3/Iris Pro GPU, the transistor count is probably around 2 billion as well.

This graph shows the much larger improvements in size and power consumption from TSMCs 16nm FinFET process

The transistor count is even more impressive when you factor in the A8’s die size, which is just 89mm2 — down from the A7’s 102mm2.

The shift to TSMC’s 20nm process certainly explains some of the increased transistor density, and also some of the power efficiency improvements reported by Apple — but certainly not all of the improvements. TSMC itself says that its clients should see 15-20% improvement in both die size and power consumption when going from 28nm to 20nm. There will be additional gains when TSMC transitions to FinFET at 16nm — but that’s not for another year or so.

The only sane assumption seems to be that Apple significantly rearchitected the A8, both to squeeze in more transistors, and to improve overall efficiency. Or Apple just threw in a lot more L2 and L3 cache, which costs a lot of transistors. Specifics will have to wait until a reverse-engineering company like Chipworks actually gets its hands on the A8 SoC and takes a look inside.

iPhone 6 A8 SoC CPU and GPU performance graph (Apple’s own figures)

CPU and GPU specs

In classic Apple style, the company only revealed the bare minimum of tech specs for the CPU and GPU inside the A8 SoC, instead falling back on relative performance graphs. As you can see from the graphs above, the A8 CPU and GPU are about 25% and 50% faster than the A7 CPU and GPU.

25% faster on the CPU side means we can almost certainly discount an increase in core count. Instead, we are probably looking at a tweaked dual-core Cyclone CPU (higher IPC) that’s clocked slightly higher (perhaps 1.4GHz instead of the A7’s 1.3GHz). If you recall from last year, Cyclone is Apple’s 64-bit ARMv8 chip; it’s a superscalar beast that has more in common with desktop CPUs than something like Qualcomm’s Snapdragon.

Block diagram of PowerVR’s Series6XT GPU

The 50% faster GPU performance is probably something as simple as more cores. Last year’s A7 had a quad-core PowerVR Series6 G6430 GPU. The A8 probably has a hexa-core Series6 G6630, or perhaps a brand-new hexa-core Series 6XT GX6650. A newer GPU would make sense, given Apple’s recent trend towards new, advanced graphics technologies (Metal et al.) In both cases, jumping from four to six cores would give the 50% performance boost that Apple is claiming.

A leaked iPhone 6 logic board, showing the A8 SoC and a part number that says there’s 1GB of DRAM stacked on top.

RAM and other bits and pieces

Rounding out the A8 SoC’s hardware specs, leaked Geekbench scores(Opens in a new window) and an iPhone 6 logic board appear to confirm that there will be just 1GB of DRAM stacked on top. It’s possible that there will be multiple variants — perhaps a 1GB A8 SoC for the iPhone 6, and a 2GB version for the iPhone 6 Plus. We’d just be guessing, though — which is silly, as we’ll find out for certain on September 19 when people start benchmarking their new Apple smartphones.

Given Apple’s claim of a 50% improvement in power efficiency, and only 15-20% of that coming from the 20nm process, there could be some other interesting tweaks to the A8’s overall architecture. Unfortunately, until we actually play with a new iPhone 6 or iPhone 6 Plus , it’s very hard to reveal what those tweaks might be. We could be looking at improved power gating — or maybe tweaks to the CPU and iOS 8 allow for a faster race-to-idle.

In any case, we should know a lot more about the A8 SoC after the

In any case, we should know a lot more about the A8 SoC after the iPhone 6 and iPhone 6 Plus are actually released on September 19 . Reverse-engineering companies will tear the chip apart physically, developers and hackers will interrogate it mentally. Rest assured that the A8 SoC will give up all of its secrets sooner rather than later.