If there's one thing that's consistent about Apple (AAPL)'s smartphones and tablets, it's that the Cupertino giant has always been obsessed with providing the best performance that it can. This is why the company has steadily bolstered its chip design teams over the years, releasing successively more impressive chips with each iOS device. Over the last six years, Apple has built a world-class semiconductor operation and its efforts here are likely to continue to impress.
Apple Likes to Brag
Apple said that the A7's graphics engine (which is widely believed to be an Imagination Technologies (IGNMF) PowerVR G6430) is, according to Apple, roughly twice as fast as the SGX543MP3 found in the previous A6 chip. While this performance uplift will vary based on the particular workload, the point is that the increase is fairly substantial. And, since Apple enjoys putting up these pretty charts showing how much better the new chip is over the prior generation, it will need to deliver a similar jump with the A8 — if possible.
Advances in semiconductor performance can be achieved in a variety of ways. Barring a transistor/manufacturing process shift, performance can be achieved by raising the power envelope or making the chip bigger — both of which are trade-offs that need to be made by the chip architects based on what product that chip is going into. With a manufacturing technology shift (and in the A8's case, a move from 28nm HKMG to TSMC (TSM) 20nm HKMG), the transistors can operate faster at a given power level or they can operate at the same performance level in a lower power envelope.
So, depending on the dollar budget for the chip, the move to 20-nanometer means that the company can either:
Spend the added area headroom in order to increase performance by simply using more of the transistors for graphics.
Reduce the area but then take advantage of the added thermal headroom engendered by lower power transistors to bump up performance (i.e. boosting the clock speed).
In this case — since Apple's margins are so high on the iPhone and the chip is such a relatively small part of the bill of materials — it would not be a surprise to see Apple use the smaller transistors in order to pack a lot more of them in the same area (even if this drives cost up slightly). Indeed, as you'll see in a moment, this is Apple's only option to really drive performance up.
A Move to Imagination's G6630 Or GX6650
According to AnandTech, here is a list of the various members of the PowerVR Series 6 and 6XT IP cores from Imagination (Apple's GPU partner):
The iPhone 5s uses the G6430 which is the second fastest member of the Series 6 family, so the next logical progression would be to use either the G6630 or — if Apple was ninja enough to implement it so quickly — the GX6650. A move to the G6630 would improve performance by roughly 50% at the same clock speeds (at the cost of area and power against the G6430 on the same node), but in moving from the 28nm node to the 20nm node, Apple would have the area budget and lower power/higher performing transistors to work with.
Now, if Apple were able to implement GX6650 and if the GPU could be very aggressively power/clock gated, it could potentially swing another 2x performance improvement over A7's GPU. Further, if this chip is destined for a 4.5-4.7" iPhone, there could potentially be a greater thermal envelope to work with. As a result, Apple may be able to more aggressively clock a G6630/GX6650 block in the iPhone 6 chassis than it would be in the iPhone 5s chassis. In short, it would be unwise to assume that the big gains in graphics performance have come to an end. Imagination should benefit from a move to higher value IP.
A8 Will Be a Monster
Thanks to a move to TSMC 20nm, an updated SoC micro-architecture, improved IP blocks all-around, and a large budget by virtue of the high margins and importance of the iPhone to Apple's business, the iPhone 6 will be a performance monster, particularly in graphics. Longer term, Apple will probably want greater control of the graphics IP in its SoCs and will either roll its own GPUs or it will push Imagination harder and — by virtue of being one of Imagination's largest shareholders — could be granted some level of early access to the latest IPs. Indeed, it's no surprise that Imagination has been touting an accelerated IP roadmap at its various investor conferences, and it was the first to ship a commercial implementation of PowerVR Series 6.
At the end of the day, I believe the iPhone 6 will be able to spit out some pretty serious graphics performance thanks to both improved designs/micro-architecture as well as a move to next generation manufacturing technology. This is the biggest single node jump that Apple's chip team has had since the A5->A6 transition, so the performance/watt improvements across the board should be much more dramatic in moving from A7->A8 than we saw in moving from A6->A7 which saw a half-node move from 32nm HKMG to 28nm HKMG. The downside though is that Apple (and the rest of the ARM ecosystem) did substantially drive up power consumption, a one-time benefit that will not be repeated in future generations.