The Intel-ARM War Heats Up

So the war between Intel and ARM has just heated up a notch, and things are really starting to get interesting. Intel just showed off the first Android smartphone powered by an Intel Chip, which was clearly a warning shot in ARM's direction. Performance is allegedly better than the current dual-core ARM designs, although no one has actually been able to play with the device yet. Power consumption is supposedly comparable to existing dual-core ARM devices, although we won't know any of this for sure until we see production designs.

More importantly, Intel announced high-profile partnerships with Motorola and Lenovo, who have claimed enthusiastic support. Within the next six months or so, Intel devices will be available at your local smartphone store. And Intel is committing to significant improvements to their technology over the next few years.

What does this really mean? Is Intel aiming to beat back ARM's advance into its territory (tablets and notebooks), or is it trying to capture the smartphone space from the current leader? ExtremeTech's Sebastian Anthony just published a column alleging that Intel will dismantle ARM. He claims that Motorola's "extensive" partnership provides proof that Intel has something special up its sleeve. Intel's process shrinks over the next several years will allow Intel's chips to gradually outpace ARM's designs. By 2014, Intel's mobile chips will be the clear leaders, and ARM might as well pack up and go home.

What Actually Happened
Well, I'm not so sure I agree, so here's what I think. Intel was scared when they heard that Windows 8 would be running on ARM. Sure Intel had lost its foothold in the mobile space, but it had pretty solid dominance over the desktop and laptop market. Windows 8 running on ARM is a slippery slope. At first, it may just be for tablets, but tablets are already starting to include HD displays and clip-on keyboard, which makes them functionally equivalent to laptops. If they didn't make a move, Intel knew that they would be disrupted, gradually pushed upmarket and eventually obsoleted completely.

So Intel basically approached two companies who are down on their luck and offered them sweet deals to use their chips. I really can't say much about Lenovo because their devices aren't available stateside, but Motorola has made many errors in executing its smartphone and tablet strategies. Motorola's tablets were a bust, and based on the recent financials, their smartphone sales are stalling again. They were looking for some special sauce to differentiate themselves, and along came Intel.

Enter The Same Old New Thing
In coming up with a product, Intel did what they always do in this sort of situation. They pulled out old technology, and spun it as new. Here's a secret - the processor in Medfield is functionally identical to what Intel released in 2009. It's a single-core, dual threaded Atom processor running at 1.6ghz, exactly the same as you would find in a first generation netbook. The only major difference is that it is significantly smaller, and that a bunch of other chips are integrated onto the CPU (reducing the number of support chips required). And the Atom was essentially five year-old technology when it came out - it was essentially a Pentium M (circa 2003) that was optimized to use less power.

Atom hasn't increased in performance in the past three years. All Intel has done is to shrink the die, reducing both physical size and power consumption. At some point, the size and power consumption both dropped to the point where you could fit an Atom processor into a mobile phone. And heck, that Atom processor's performance is slightly favorable to the current generation of mobile phone chips. So Intel might as well use it.

Intel is a One-Trick Pony
But here's the problem - Intel doesn't have any more tricks up its sleeves. It just happens that performance of Medfield is comparable to the competitions, because the competition has steadily improved as Intel has stagnated. Medfield is Atom from 2009, but Intel hasn't significantly improved Atom since then. They just focus on their higher-end (and higher margin) chips, and trickle down the process improvements to Atom. They were so afraid of cannibalizing sales of their higher-margin products that they intentionally crippled their lower-margin ones. Sure Intel can go to a dual-core design (as Atom eventually did), but there are no magical architecture improvements on the horizon.

The only real improvement that Intel can promise over the next two years are die shrinks.  And die shrinks are a great thing, but they don't solve all of the world's problems. First of all, die shrinks are somewhat unpredictable. To do a die shrink, you usually have to build a new factory with all-new equipment. Factories often take longer to build than expected, and frequently the initial chips don't work nearly as well as hoped for. Typically, you only do a die shrink on an existing design - changing both manufacturing process and design at the same time is a recipe for disaster. If Intel hopes to die shrink twice, first to 22nm and then to 14nm, they are going to have to stay with essentially the same Atom we currently know and love. Given that 14nm is about 1/4 the size of 32nm, you will probably see 4-core Atom designs where we currently have one core. In fact, Intel's next generation of Atom is nicknamed Clover Trail, which indicates that it will likely be a four-core solution (designed to take on the ARM-based Windows 8 tablets). With a few die shrinks, it will fit in your phone.

Why ARM Will Win
So here's the thing about ARM that has caused the platform to advance so quickly. All that ARM does is design chips - it doesn't make the finished product. That falls onto TI, NVidia, Qualcomm, Samsung, and Apple, who all make chips based on the ARM designs. These companies don't design the chips from scratch, but they do tweak the designs significantly. Right now there's a massive war between those manufacturers to produce the fastest and lowest-powered chips. On average, each company releases a new chip generation at least once a year, and these lead to huge improvements year-over-year. By then end of last year, Tegra2 (which came out in January) had been eclipsed by TI's OMAP4, to the point where no new smartphones include Tegra2. The quad-core Tegra3 is both lower-power and higher performance than the Tegra2, and should blow the pants off of the first-generation Intel smartphones. By the time Medfield is released, Tegra3 should be included most high-end devices. By next year, when Intel releases the first die shrink (probably with a dual-core Atom), we may see the first 6 or 8-core ARM designs.

Furthermore, ARM-based designs are truly created for mobile use, unlike Intel's chips, which are simply repurposed desktop designs. Desktops and laptops are essentially either on or off at any point in time, and waking from standby takes a few seconds, even on my Mackbook Pro with SSD. On the other hand, smartphones and tablets need to operate in a low-power standby state, with the ability to instantly come to life. In response to this, ARM designs include innovative features, such as Tegra3's low-power core, which handles standby processing at a fraction of power. Since ARM licensees are developing mobile chips full-time, they will continue to create new designs that include these sorts of features.

Sorry to say it, but if Intel had really wanted to compete with ARM, they would have already done it, and this move wouldn't be a knee-jerk reaction. What we are seeing now looks just like disruptive innovation, and Intel is making what amounts to a last gasp effort to compete. Otherwise, Atom would be well on the way to disrupting Intel's midrange, and there would even be server-level Atom chips.

So here's what I predict is actually going to happen. Motorola and Lenovo are going to release Intel-based smartphones. They are going to be sold at a premium price, and will be slower than contemporaneous Tegra3 designs. Overall, they will be a flop, and the manufacturers will eat crow. Motorola Mobility, which will be owned by Google, will return to producing primarily ARM-based designs, keeping one Intel-based design around for show. The second generation Intel processors will be more competitive, but no one will really care by that point (they will probably work their way into some low-power tablets and notebooks). ARM will continue to improve - I think Nvidia's road map indicates that the 2015 model will offer 100X the performance of Tegra2. In 2014, or 2015, we will see the first high-performance ARM-based laptops. By that point, Intel will be relegated to server duties.

Regardless of what happens, I think that the consumer will win. More competition leads to greater innovation, and we are currently seeing unprecedented levels of advancement (the last time we saw anything of the sort was the late 90s/early 2000s, when ATI and NVidia were releasing new graphics cards every six months. The only reason Intel has had to improve their chips so much is that formidable competition arose from an unexpected place. If not for that, Intel would have reserved their next few die shrinks for their higher-end processors, and Atom would get them a few years later (if ever). This is definitely going to be an exciting time for mobile devices.