Apple sells devices. Lots of them. Their success is due to many things related to design and tech religion, and an important part is the SoC inside those devices which creates the experience people want. The official Apple information on their parts is minimal. Their SoCs have been dissected with more fervor than Roswell aliens. We know some but not every detail, but connecting the dots of some history tells more of the story.
To understand the Apple approach to today’s A5X SoC, one has start all the way back in 1987. The search began for a processor in a new type of device, the “personal digital assistant”. Apple spotted ARM after experimentation with the AT&T Hobbit, and eventually became an investor when ARM originally went public. After a few misstarts and the revolving door drama of management and strategy changes, the Newton released in 1993 powered by the ARM610, a 32-bit 33MHz workhorse consuming 500mW. The Newton didn’t quite hit the intended target, perhaps being a few years too early and a bit short in usability and a bit big in size.
But the seeds were planted for the design of the iPhone and iPad. As Apple’s design thinking evolved, the ARM community also kept innovating in parallel, making faster and smaller things possible. While Apple uses the same primary ingredients as most of today’s mobile SoCs, starting with the ARM Cortex-A9 core, their implementations are unique due to (at least) two EDA technologies.
Dan Dobberpuhl’s P.A. Semi took their StrongARM experience into the Power Architecture market for a while, creating the super-efficient PA6T processor which debuted in 2007. Over 50,000 clock gates in the PA6T meant instead of shutting down whole cores and functions, they incrementally turned on only what was needed for the operations underway. This was compelling for Apple’s vision, and a lot of heads turned when P. A. Semi’s 150 employees were purchased and integrated into Apple in 2008.
Also in the Apple stable is Intrinsity, a small firm just outside of Austin that specialized in a unique multi-level logic and multi-phase clock EDA approach that could drastically reduce the wires, transistor count, size, and power of execution units inside a chip. To prove out the technology, Intrinsity collaborated with Samsung on an ARM Cortex-A8 core implementation. They were widely coveted, and Apple quietly purchased Intrinsity in 2010 for this capability.
Apple’s current SoCs are the A5 inside the iPhone 4S and iPad 2, and the A5X inside the new iPad (aka iPad 3 unofficially). If you haven’t seen it, you might want to look at the Chipworks new iPad teardown.
CPU: both the A5 and A5X feature the dual-core ARM Cortex-A9 running at 1.0 GHz, and a dual channel LP-DDR2 memory interface running at 800 MHz. The A5X remains for now on the 45nm Samsung process, while the A5 inside the iPad 2,4 has been process bumped (but not speed bumped, trying to reduce costs) to a 32nm Samsung process. Again: the iPad 2 and the new iPad have exactly the same dual 1.0 GHz Cortex-A9 inside (the iPhone 4S is backed off to 800 MHz).
GPU: the A5X improvements lie in the graphics. The new iPad is all about driving a huge number of pixels, 2048 x 1536 compared to 1024 x 768 on the iPad 2. That task in the A5X goes to the Imagination Technologies PowerVR SGX543MP4 running at 250 MHz. That’s the same clock as the SGX543MP2 in the A5, but twice the execution units to scale to twice the display.
Modem: Our last installment triggered discussion on to integrate or not integrate baseband, so I thought I’d mention Apple uses an external Qualcomm MDM9600 with the A5X in the new iPad. So much for the thought of “must have” integrated baseband for devices to be successful.
On today’s landscape, 1.0 GHz is relatively pedestrian for a mobile SoC, but the Apple magic keeps up or even outpaces just about everything dual-core Cortex-A9 out there. A hot debate right now is the A5X versus the NVIDIA Tegra 3, and I’m a bit skeptical to show results here since they are under a lot of dispute. CPU benchmarks that thread onto multiple cores fall well in favor of the Tegra 3. GPU benchmarks are at least initially falling in favor of the A5X.
I’ve been questioned what is “better”. The A5 and A5X are clearly fast and nearly equivalent in CPU performance benchmarks. If one is looking for a fast device but not the highest video performance, the iPad 2,4 with the process bumped A5 might be the best value. If one is more concerned about a fast, high quality display, the new iPad is outstanding (but our own Daniel Payne suggested that the web standard 72dpi graphics are actually a bit fuzzy on it, too low res). If one is looking for super CPU multitasking and a lot of apps running at once, something quad core may be more in order, but that comes at a price of power consumption.
The Apple A6 is in the wings, presumably for the long awaited iPhone 5 later this year, and it’s highly speculated as to what’s inside. Consensus seems to indicate the A6 is going quad Cortex-A9 and PowerVR Series 6, and it’ll be fabbed at TSMC, or Samsung, or even Intel based on various reports. I’m going to stay out of that speculation here, I can guarantee I’ll be wrong – but I’d love to hear informed comments from my counterparts and others.
Apple’s allure is in the overall device experience, and it’s a good bet they’ll work hard to maintain their lead with users, no matter what exactly is inside.
Next up in the Smart Mobile SoC series: NVIDIA.
(Disclaimer: Not sponsored by, or invested in, companies mentioned. Never free, never me, but I am inexpensive when you need help with content on topics in social computing like this.)