TWO NEW JOBS posted in December on Apple’s Mac Hardware Engineering jobs website strongly point towards a big change coming in Cupertino.
Steve Jobs’ team are looking to bring onboard a new CPU micro-architect and implementation engineer. To the uninitiated, that may not sound like a big deal, but these roles are very specific and can only mean one thing: Apple is designing their own CPU from the ground up.
This is what many have suspected for a while and is a further continuation of their vertical strategy putting them right into competition with Intel and Qualcomm.
But the biggest news is that, with the job placed in the Macintosh jobs section rather than just the iPhone, what is potentially at stake here is the transition of Macs, presumably starting with laptops, to ARM, which would be a huge shift for the computer industry in which x86 has been dominant for so many years.
This is all somewhat complicated so here’s a bit of a history lesson. For the first iPhone, Apple sourced a SoC (System on Chip) from Samsung who, in turn, licenced a synthesized (pre-designed) core design from ARM – an ARM11, specifically. With the iPhone 3GS, no suitable synthesized core was yet available from ARM for the Cortex-A8 so Apple and Samsung brought in a company called Intrinsity who had designed a speed-optimised Cortex A8 based on the ARM VHDL code. Apple acquired Intrinsity, bringing this in-house.
Moving onto the iPad and the iPhone4, Apple finally got round to using the engineers they had earlier brought in from PA Semiconductor to design the SoC themselves based around the Intrinsity core design, reducing Samsung to a pure manufacturing partner for the new A4 processor.
However, ultimately, one major piece of the puzzle remained out of Apple’s control: their use of ARM’s Cortex A8 architecture. Sure, they could choose the A9 or A15 but so could everybody else. Own your own team and throw cash at it and you can do better.
You can only optimise the design so much when it is someone else’s design. The Cortex A8’s pipeline is inherently of a fixed length, it is out-of-order and there are many other key characteristics which Apple just can’t change. The use of a standard core which is available to all their competition limits their ability to differentiate on this important element of the hardware design and even permits companies to one-up them by doing their own so, naturally, it is a target for Apple’s vertically-integrated business model.
To design your own ARM-compatible CPU (not just take their design), you need an architecture licence from ARM. Companies such as TI, Broadcom and Samsung take the core from ARM’s team but Qualcomm and Marvell design their own architectures, under licence, compatible with all existing ARM software but which may be faster or slower at a certain clock speed and use more or less power. The architecture licencees are like the AMD and VIA to Intel (except they pay ARM and didn’t have to reverse engineer anything).
It’s probably helpful to clarify all this a little further. Look how Apple’s ownership of the processor has increased with time:
Combined with the fact that the team from PA Semi already have considerable expertise in the CPU micro-architecture area (having designed the first ever PowerPC core outside of the AIM alliance – Apple, IBM, Motorola), this looks to be inevitable. PA was founded by Daniel Dobberpuhl, designer of both the StrongARM and Alpha processors at Digital.
However, while there would be many unsurprised by Apple designing their own competitor to ARM’s Eagle (the Cortex-A15) for the iPhone 6, it’s a lot more interesting to spot that ARM may finally escape from mobile devices and shake up the personal computer market which has been pretty much exclusively x86 since, well, Apple announced their shift to Intel and x86 in 2005.
Not that it’s been for a lack effort that cheap ARM notebooks have failed: plenty have had a go. Dell even created a hybrid Intel/ARM laptop. But all found that consumers expected that something which looked like a PC (and wasn’t an Apple) would run Windows and their existing apps and would pay a bit more to get that.
Furthermore, despite ARM’s protestations, the raw performance of their chips simply haven’t been able to match even the slowest x86 chips that Intel made until 2009.
Bizarrely, the decisions Apple made back in 2005 when moving to x86 may well now help to reverse that momentous decision as, with universal binaries and Apple’s increasingly competent development tools, it should be easy to create an app that targets both architectures. Let’s also not forget that, back in 2005, there were no third party apps made for Mac OS X running on Intel before the announcement, hence the need to make the “Rosetta” emulation layer to run legacy apps.
Now, whether even a future 2GHz+ dual-core ARM chip would be capable of running heavy software like Photoshop CS5 in an emulator like Rosetta is highly debatable but one would imagine that there are a large majority of users who never go anywhere near software like this. Apple now have their own software for the majority of office and home tasks that they could recompile and an increasing number of things can be done by web apps that need very little from the PC itself. Suddenly, backwards compatibility isn’t the big deal it used to be.
Oh and there is that small matter of the 300,000+ iOS applications which, form factor and UI constraints permitting, could run straight away from the new Mac app store without even recompilation. Also, how many more developers are there now than 20 years ago who could create that new killer app?
So the times, they are a-changing. and Apple makes the sensible argument that performance is becoming less reliant on the speed of the CPU and more on the graphics, mass storage, Internet connection and battery life. Reviews of the new MacBook Air point to snappy response times despite their use of a ridiculously slow 1.4GHz version of the 4-year old Core 2 Duo. This is a prime case of putting performance where it’s needed rather than just trying to boost the headline figures so semi-geeks will “understand” that it’s “better”.
With the staggering growth of smartphones and tablets as Internet devices and loud rumours of Microsoft imminently announcing a mainstream version of Windows 8 which runs on ARM, Intel’s management is already undoubtedly feeling somewhat nervous. This news is unlikely to improve the situation.
Designing a new SoC based on your own microarchitecture, however, is not an overnight task. It’s a fair assumption that this is something which has been worked on since soon after PA were acquired in 2008 but it won’t have taken them long to start. Conservatively, you’d expect such a chip to launch in the 2011-2012 timeframe on a 28/32nm process.
But one thing’s for sure: manufacturers have something big to gain introducing ARM-based products: high-end ARM CPUs tend to cost in the region of $15-25. If Apple designs their own, it could cost even less. Assuming this chip finds its way into the MacBook Air first, it would replace the Core 2 Duo SU9400 which costs $262 and requires an nVidia chipset in the region of $60-80. That’s a huge saving whichever way you look at it. Add in the massive savings in space and you have a very compelling argument. Who cares about Centrino branding now?
It’s still the king, no doubt, but x86 is now under its first real attack in years and Apple and Google look to be leading the charge with some enthusiasm.
Intel’s engineers are at their finest when under attack from competitors whose products have actual advantages so it would be crazy not to think they could come up with a strong response. Look at how they demolished the Athlon 64 with Conroe. Likewise, AMD’s new Fusion platform looks to ARM’s “just good enough” CPU performance paradigm, backing it up with top notch graphics and shows great potential too.
Perhaps the outcome will be that the future world of computing will be a lot less architecture-dependent than today’s. But, for the man on the street, a shaking up of the industry giving more competition means better battery life, higher performance and lower prices which should be great news for all of us. And it’s brilliant news for ARM, who stand to win big time by any advances into the world of PCs.S|A