Another rumor just hit the news wires today, courtesy of RBC Capital. I caution my readers to, of course, take any and all rumors with a grain of salt. A few weeks ago, the buzz was about Apple (AAPL) dropping Intel (INTC) chips from its iMac and MacBook Pro products. Today, the rumor is a complete 180 degree turn from that pessimistic view, claiming that Intel and Apple are in talks for an elaborate deal that would consist of the following:
- Intel acts as a foundry to Apple to build its A-series of mobile processors for the popular iPhone
- In return, Apple uses Intel designed-and-built chips in its iPad.
On the surface, we can see why this deal would be mutually beneficial. Right now, Apple has its custom-designed A6 and A6X system-on-chip solutions built by Samsung, since it is -- other than Intel -- the only company in the world that can guarantee the volumes and manufacturing space that Apple needs. However, tensions are high between Samsung (OTC:SSNLF) and Apple, especially as lawsuits over software and physical design patents seem to be escalating to brand-new highs each day. Reportedly, Samsung allegedly increased the price of manufacturing Apple's A6 processors to Intel by 20%. For Apple, a company that is hit with fears of gross-margin compression, this is unwelcome.
Apple likely wants out of the fabrication deal that it has with Samsung, and it needs to find a partner that's as reliable, and who isn't a direct competitor. Enter Intel.
Apple Fabbing At Intel Makes A Lot Of Sense
Intel designs, builds, and sells chips. It doesn't matter who buys the chips, and it doesn't matter which devices they end up in, as long as Intel is selling them. In addition to having the best micro-processor designs for everything from ultra-thin laptops to the biggest and baddest dataceneters, it also sports the best manufacturing technology in the world. While the majority of Intel's competitors -- Taiwan Semiconductor (TSM) is the most advanced one -- are still struggling to add 28nm capacity, Intel is idling 22nm capacity and moving some of its older fabs to its brand-new, next-generation 14nm process technology.
In such a potential partnership, Apple would be guaranteed to avoid shortages of its latest applications processor for its iPhone in addition to access to leading-edge process technology. Intel has more capacity than it can fill currently (utilization in Q4 is sub 50% to bring inventories down), so this works out for both parties economically. In the low-power space, Intel currently ships a 32nm HKMG process, which puts it in the same class as the foundries (which are generally shipping 28nm either with or without HKMG at the bleeding edge). Intel's process roadmap on the low power space is being accelerated to be available at roughly the same time as the higher performance processes:
This means that in a couple of generations, likely in the 2014 time frame (when the Samsung/Apple contract is set to expire), Intel will offer a tangible process lead over the rest of the foundry industry. With smaller geometries, SoC designers can either produce smaller and cooler chips than the competition at the same transistor count, or produce superior and more featured designs in the same area. This is an advantage that cannot be overstated. The best design in the world could always be better if it had a bigger transistor budget and/or lower power per transistor.
It also goes without saying that since Intel is not a direct competitor to Apple on devices, there is much less of a conflict of interest than there is with Samsung building Apple's chips.
Apple Should Want Intel's "Tablet" Chips
In such a deal, Intel would benefit greatly. While acting as a foundry to Apple for the smartphone processors is nice business, the real jewel would be an Intel chip in the iPad. Intel is very serious about being in tablets/convertible laptops, so scoring the iPad with its processors would be a noteworthy win for both its Atom and Core based processors. With the more performance-oriented iOS device running on Intel's chips, along with the Apple software ecosystem extended to include Intel Architecture, Intel would have the bulk of the personal computer industry.
Now, I am keenly aware that Apple's A6X ARM-compatible chip is a quite powerful upscaling of the iPhone 5's A6 SoC. However, a key trend to notice is that "Ultrabook convertibles" and eventually high end "tablets" will be running on Intel's high end "Core" processors rather than the budget-oriented "Atom" processors that compete with the ARM (ARMH)-based designs. At the Intel developer forum, Intel demonstrated its next generation "Haswell" CPU, running at a power-consumption level of merely 8W (compared to 17W in today's lowest power "Ivy Bridge" Ultrabook processors).
To put this in perspective, Anandtech tested the latest-and-greatest Samsung Exynos 5250 based on a dual-core "Cortex A15," ARM's finest CPU core design, which is faster than Apple's already fine A6/A6X design. The power results were rather shocking:
At idle, the Cortex A15-based machine consumed 6.3W. At full CPU load (remember: the SoC's TDP also included graphics processors, which would introduce additional power usage in full "real" load), the chip consumed an average of 10.5W, or 4.2W. So we can safely estimate that the Exynos 5250 for Chromebooks/tablets has a TDP of 5W. Still lower power than the up-and-coming Haswell. However, this is not the end of the investigation.
The Linux-oriented website Phoronix recently published benchmarks of the Chromebook running against the Atom D525 (it is equivalent in speed to the Atom Z2760 in tablets, but in a higher power envelope since it is older), a Core i3 330M, and an Nvidia (NVDA) Tegra 3.
Now, to add some color here, the Core i3 330M is a 32nm dual-core design based on the now old "Westmere" CPU core (two generations behind "Ivy Bridge," and three generations behind "Haswell") running at 2.13GHz. The chip also has a 35W Thermal Design Power or "TDP."
The benchmarks, of course, generally consisted of graphs that looked like this:
The Atom processor from 2010 (based on a 2008 core design) being slower-to-on-par with the dual Core A15 (as expected). However, the older "Core" design regularly proved itself to be multiple times faster than the Exynos design. Granted, this is at a much higher power envelope. 35W versus 5W should not even be allowed, right?
Well, here's why it's permissible. At that IDF demo, the 8W "Haswell" was shown to be on-par in performance to the 17W "Ivy Bridge". Now, a quick look at the specifications of the 17W "Ivy Bridge" reveals the following:
In the 17W power envelope, at the 22nm process node, we have a 1.9GHz base clock, dual-core/four-thread "Ivy Bridge." On a per-clock basis, the "Ivy Bridge" is faster, since it is based on a newer, more sophisticated micro-architecture, and the two designs run at roughly the same clock speed. Now, imagine a 10W or less iteration of "Haswell" with roughly the same performance characteristics of the 2010 Core i3 part in the benchmarks.
This would imply that in the 10 watt (or less -- Intel referred to the ultra-low power part as "sub 10-watt" at the Credit Suisse Conference), Intel would have a very significant performance per watt advantage over the latest-and-greatest ARM design. The lead should only widen at the 14nm node with "Broadwell" as lower power levels are enabled.
If Intel is able to bring its very-highest performing processors to the tablet power level, then Apple will have very compelling reasons to move its iPad to Intel designed and built processors. But there is one snag -- software.
Software Compatibility A Big Problem, But Not A Showstopper
The big thing that would keep Apple from moving to an Intel chip, even if it were superior, is software compatibility. iOS runs on ARM-compatible processors, and all of the apps for iOS are compiled for and run on ARM-compatible chips. Now, moving most of the applications would be tough, especially since iOS developers do not submit source-code, but simply finished executables. On Google's (GOOG) Android, Intel caught a break because almost all of the code runs via the Dalvik Virtual Machine (meaning if the OS supports the instruction set and has the proper virtual machine, then the applications will work seamlessly).
Nothing stops Intel, however, from using the same kind of software-voodoo that it did with native (i.e., compiled) Android applications: binary translation.
Intel could provide a similar tool for iOS products. In fact, since the high-end Intel chips would have a tangible performance advantage over the ARM-based products, the penalty felt from the overhead involved with binary translation for existing applications would likely be nonexistent.
There are technical hurdles, but they can be overcome.
I have no idea if this rumor is true or not. However, with Intel moving its very highest performing processors to lower and lower power levels, thanks to a combination of a sophisticated process as well as a highly optimized design, it is likely that Intel will attack the high end tablet/convertible space with "Core" and attack the low end of the market as well as phones with the new "Atom" architecture. If true, then this sheds significant light on the fairly delayed introduction of the next generation "Bay Trail" platform based on the next-generation "Silvermont" Atom core.
Should Intel actually capture the iPad with its Atom and/or Core products while filling its fabs with Apple A-series chips for the iPhone, this could represent very real and significant upside for shares of the company.
Additional disclosure: I'm short ARMH