At Hot Chips this past week, Intel's Rajesh Patel revealed more details of the chipmaker's upcoming Moorestown platform. Patel, who is the lead architect of the platform's system-on-a-chip (SoC) part, also made it clear that he really does believe that Moorestown is a bona fide "smartphone" platform that will give us the world's first x86-based phones. I wasn't actually at the presentation, so I didn't get close enough to Patel to tell if there was any crack smoke on his breath as he made this smartphone claim, but I did get hold of the materials, so let's take a look.
Intel still isn't giving any real power draw information on Moorestown, other than to say that the platform has reduced Menlow's 1.6W idle power draw by 50 times. There's not a word said on how much it has reduced Menlow's peak power draw, so my guess is that the answer is "not much." This being the case, it's safe to assume that Moorestown's peak power draw is still well above (possibly by an order of magnitude) a comparable high-end ARM-based solution, but Intel plans to make up for this with a combination of very low idle power and a whole ton of spatial and temporal granularity in frequency scaling.
The result will be a part aimed at what Intel calls "high-end smartphones" (think the new Nokia 900, for example), with very good performance-per-watt for low-intensity smartphone workloads. But what will probably keep this hyperthreaded, x86-based beast of a "smartphone" platform out of any real smartphones is the fact that it has way more absolute performance than anyone needs on a phone right now. Intel will be asking users to pay for that performance with worse battery life than a comparable (probably single-chip) ARM-based solution, when in the end the ARM part is perfectly adequate for 99 percent of what people do on a smartphone, i.e., casual gaming, media playback, Web surfing, messaging, social networking, etc. I think almost everyone is going to pick the battery life over the performance.
Moorestown will be a better mobile gaming platform than anything from ARM, though, so if someone makes a gaming handheld out of it that won't be a bad thing. It's also the case that the platform may have a great future in point-of-sale terminals and other verticals where x86 compatibility is actually an advantage.
Ultimately, though, I expect that in late 2010 I'll be writing an article about the channel availability of Moorestown-based smartphones that reads a lot like the one I did on Menlow-based MIDs. The real x86 smartphone action is going to happen at the 32nm process node, with Medfield, Moorestown's successor. But that's a topic for another day.