Columnists

The Intel Developer Forum: ARM-wrestling

Analysis: Intel's strategy for handheld communications-plus-computing takes aim at ARM and aspires to deliver a true Internet experience to pocketable devices.

By Brian Dipert, Senior Technical Editor -- EDN, 9/21/2007

Back in late June of 2006, when Intel announced the sale of its ARM-based XScale product line, a divergence of opinion emerged among EDN's staffers as to what this move signaled regarding the future of Intel's handheld computing and communications efforts. As I wrote a few months later (and week shy of one year ago) at the fall 2006 IDF:

My theory was that Intel wasn't giving up completely on the category but would revisit past history and, in competition with the ARM approach it formerly embraced, advocate a single-core, ultra-low-voltage spin of its Core x86 architecture.

At that IDF, I tried to pin down Intel's Anand Chandrasekher, senior vice president and general manager of the Ultra Mobile Group, on the specifics of the company's plans. I received only generalities in response, but Chandrasekher "telegraphed" his messaging enough that I strongly suspected I was on the right track.

So did I nail Intel's plans 15 months ago? Witness Silverthorne, in a foil from a Wednesday morning keynote by Dadi Perlmutter, Intel's senior vice president and general manager of the Mobility Group. Silverthorne is one of Intel's many Penryn-based and 45-nm-fabricated products scheduled to ramp into production over the next year.

It's "a single-core, ultra-low-voltage spin of the Core x86 architecture," exactly as I predicted, paired with a chipset called Poulsbo in a platform called Menlow, and slated to hit the market some time in the first half of 2008. Chandrasekher, who followed Perlmutter, reiterated and expanded on the earlier Silverthorne mention, with eyebrow-raising quotes such as:

This is a...45-nm, 12-in. wafer featuring Silverthorne on it. There are a couple of thousand Silverthorne die on this particular wafer.

He also provided a teaser reference to the follow-on Moorestown project, which I strongly suspect will be a single-chip integration of Silverthorne and Poulsbo. Moorestown will be akin to what Intel's got planned for its Timna-concept-resurrecting Tolapai, albeit in this case in a MID (mobile Internet device)-optimized feature-set fashion.

The primary selling pitches of Intel's x86-in-handhelds push, as repeatedly echoed in both Perlmutter and Chandrasekher's keynotes that morning, their joint post-keynote Q&A, and the post-Q&A mobility briefing later that morning, as well as Paul Otellini's keynote the prior morning, include:

1. A rich Internet experience, versus the neutered and otherwise nonmainstream version of the Internet delivered to most handsets today via approaches such as:

• Stripping out Flash and Java streams in conjunction with server-side mobile browser detection
• Employing the .mobi top-level domain for delivering custom mobile-tailored HTML material, and
• Using even more restrictive WAP content coding.

As Otellini said on Tuesday, "The Internet compatibility, while it can be made perfect, isn't perfect today. Any part of the Internet can be transcoded to run on any device. But that's a big job. The Internet's large. From our perspective, having Internet compatibility from the start, across the breadth of the Internet, is a really exciting value proposition. The solution to this that we see is integrating Intel architecture."

2. The ability to quickly port (if not run as-is) existing Linux and Windows applications and other content to MIDs in a write once-run many fashion that leverages existing hardware-virtualizing APIs.

3. The wealth of x86-tailored tools that developers can harness for creating new content (sound familiar, DOS and Windows veterans?)

The obvious bulls-eye target of Intel's push is ARM, which today garners the lion's share of the mobile-phone business. In that respect, the oft-overlooked second point above particularly resonates with me. Unless you've talked to a developer, you may not understand why (aside from factors such as underpowered processors and graphics accelerators, diminutive screens, and a limited UI) the games-on-cell-phones market is so small and has so significantly underperformed to expectations (of others, not me).

In spite of efforts by organizations such as the Khronos Group, developers need to craft model-specific versions of many of their apps for mobile phones, especially if those apps, as is definitely the case with games, push the hardware to any significant degree, and even (albeit generally less so) with the Windows Mobile O/S.

The advantages of x86 code compatibility aside, and keeping in mind Drucker's Rule (an emerging technology must be 10× better than the incumbent approach in order to have a reasonable likelihood of replacing that incumbent), Intel's got a tough row to hoe, especially with companies that were burned by its June 2006 ARM exit. ARM's been cultivating its presence in various markets for more than 20 years, and by now its partners are intimate with its strengths as well as the means of dealing with its various shortcomings. Mountains of code archives exist, tailored for various ARM flavors as well as diverse support ICs. And porting all that code to x86 (or, alternatively, starting from scratch) is no small feat.

Click to enlarge

Reflective of these realities, if you peruse one of the foils delivered by Gadi Singer (vice president and general manager of the Ultra Mobility Group) during the Mobility briefing, you won't find many traditional cell-phone manufacturers listed. Instead, you'll see the names of computer companies, along with OEMs and design houses that have already honed their expertise in the communications-plus-computing Windows Smartphone and Pocket PC Phone businesses.

As an aside, I confess that I find it mildly amusing how Intel's now generally calling these devices "MIDs" instead of "UMPCs" (Ultra-Mobile PCs). Perhaps, because the first UMPC prototypes (also known as "Origami" devices, reflective of Microsoft's project name for them) were non-pocketable products in the same category as Palm's abruptly canceled Foleo, Intel has decided that a new acronym is necessary to avoid limiting the products' potential via association?

Porting a code base to another CPU architecture may be no small feat, but it is possible if price, performance, power consumption, and/or other important factors provide sufficient motivation. After all, Apple's done it within the past two-plus years—twice, as a matter of fact. Back in June of 2005, the company announced the commercialization of its long-secretive "Marklar" OS X-on-Intel program as a successor to its past PowerPC efforts. And, at the January 2007 Macworld Expo, Apple famously unveiled the ARM- and OS X-based iPhone. Apple is now in a unique and enviable position, with an operating system compiled for three distinctly different CPU architectures (don't forget the PowerPC) and a Universal application-development methodology that enables a single code build to run on multiple processor families. Talk about hardware design flexibility!

To close, I concur with my compatriot, Colleen Taylor. Intel's deeply in bed with Apple, on multiple product fronts. When I heard Otellini describe on Tuesday morning the desire to present the Internet "as is" on a pocketable handheld appliance, I suddenly had a severe case of deja vu:

With its advanced Safari browser, iPhone lets you see web pages the way they were designed to be seen...

The iPhone is, in fact, the trend-setting, initial successful example of the MID concept, and Intel even showcased an eerily iPhone-reminiscent Menlow proof-of-concept demo this week.

Will a future iPhone dump ARM and be Silverthorne-based? I consider this scenario highly likely. But of course we'll never know for sure until the day that Steve Jobs and Paul Otellini get up on stage together to unveil it to the world.