jueves, 10 de enero de 2013

What do we know about Apple's processor?

Apple's A5 processor has had some long legs. Originally introduced in the iPad 2 in early 2011, the chip has since proliferated across Apple's entire lineup—versions of it have been included in the iPhone 4S, the Apple TV, and the fifth generation iPod touch. Even the A5X in the 2012 iPad simply pumps up the A5's graphics capabilities while leaving the actual Cortex A9 cores untouched. Between the new iPod touches and the $99 iPhone 4S, it looks like the A5 is going to be for sale well into 2013 and beyond.
If the A5 is any indication, then, we'll be spending the next few years with the A6 chip that Apple introduced in the iPhone 5 yesterday. While the A5 is a thoroughly known quantity at this point, Apple's tendency to avoid talking about specifications means that we know next to nothing about what makes it tick. All we have now are three vague promises: compared to the A5 in the iPhone 4S, the new A6 promises "up to" two times the CPU and graphics performance; the A6 is 22 percent smaller than the A5; and the battery life of the new iPhone is slightly better than the battery life of the 4S. Absent any actual technical data about the new SoC, we're left to do the math ourselves.

Size and battery life

The cause of the reduced size and increased battery life of the iPhone 5 is pretty easy to guess: while the A5 chip in the iPhone 4S is still manufactured on the A5's original 45nm process, the A6 is probably manufactured on Samsung's 32nm process—Apple has already used this process on the A5 in the revised iPad 2 and the most recent Apple TV (and, in all likelihood, the newest iPod touch). Using the new process on an existing design would help Apple and Samsung to work out any kinks, clearing the way toward making a new chip on the same process. Intel's "tick-tock" processor release cadence uses the same methodology to reduce yield issues and other problems that can come from trying to make a new chip on a new process at the same time.

The 32nm Apple A5 is 41 percent smaller than the 45nm version in the iPhone 4S, so a 32nm Apple A6 would have some room to grow while still remaining smaller.

According to Chipworks, the 32nm A5 is about 41 percent smaller than the 45nm version, so a 32nm A6 would have room to add more silicon while still remaining smaller in absolute size. The 32nm A5 in the revised iPad 2 also delivers a 20 to 30 percent battery life increase over the original version with the 45nm chip, so the ten-or-so percent increase in battery life for the iPhone 5 is also pretty plausible, even given the higher-performing SoC.

Increased performance

With all of that in mind, let's look at these doubled performance claims. To get this big of a jump, you can either (1) double the hardware, going from the dual-core CPU and GPU in the A5 to quad-core parts in the A6, or (2) use newer, more efficient architectures, possibly combined with higher clock speeds (increasing clock speeds enough to get twice the performance from the same architecture would probably be too onerous from a power consumption perspective).
So, does the A6 use a quad-core Cortex A9-based chip combined with a quad core GPU? Let's look at the 2012 iPad's A5X, which uses a quad-core GPU to double the A5's graphics power and drive the 2048x1536 Retina Display. Our pals at Chipworks put the die size of the 45nm A5X at 12.90mm by 12.79mm (or 165mm, total), while the 45nm A5 is just 10.09mm by 12.15mm (or 122.6mm, total). Let's assume that if Apple decided to shrink the A5X using Samsung's 32nm process, that the die size would shrink by the same 41 percent, putting the size of a theoretical 32nm A5X at about 97.35mm, or about 21 percent smaller than the 45nm A5.

Enlarge / The quad-core GPU in Apple's A5X would make it a big chip, even at 32nm.

Now, keep in mind that while Apple says the A5X doubles the graphics performance of the A5, its CPU performance is identical to the older chip. Our 32nm A5X is about the right size and it has the right graphics performance, but it doesn't increase the CPU performance the requisite amount. Thus, it's fair to say that the A6 isn't using a quad-core CPU and GPU to achieve the stated gains.
There's also another reason to go with a faster dual-core CPU than a slower quad-core design: iOS and its apps have been running on the dual-core A5 for some time and as such are already optimized for two cores, and adding another two cores would only unlock those performance gains for developers who put in the extra work.
For these reasons, I think the current scuttlebutt is correct: the A6 is going to be the first chip to market based on the Cortex A15 processor architecture, which is said to be 40 percent faster than older Cortex A9 chips at equal clock speeds. That, combined with higher clock speeds (the A5 in the iPhone 4S is clocked at just 800MHz), could easily get us twice the CPU performance while still using a dual-core design. Even if the two cores are larger than two Cortex A9 cores, they're still not as big as four Cortex A9 cores.
The same holds true for the GPU side: by our calculations, there's probably not enough room for the A5X's quad-core Imagination Technologies SGX543MP4 along with the A15 cores, given what we know about the size of the A6, but a dual-core, faster-clocked Series 6 GPU from Imagination Technologies could likely achieve the same performance while using less space than four less-powerful cores.
Many things have to be true for these assumptions to work out: if Apple is using a 28nm process instead of a 32nm process, or if the A6 is using the same 512MB of RAM as the A5, either of those factors would probably be enough to throw off our math. But given that this is a major chip revision for Apple and that it will probably be serving in not just the next iPad, but future iterations of the Apple TV and iPod touch for the next few years, the use of forward-looking technology makes more sense than would beefing up old technology with more cores or higher clock speeds.
When we know more about the A6 for sure, you can be sure that we'll take a closer look at the chip and its implications for the rest of the market and for current and future Apple devices. For now, this represents our best guess.

No hay comentarios:

Publicar un comentario