The New Intel Mobile Division

14 Comments

View All Comments

• MonkeyPaw - Wednesday, December 14, 2011 - link

  I just don't know if intel will be able to make it in the ultra low power class. Atom has not come far enough yet, and ARM is showing quad core solutions and faster dual cores today. Intel really start to shine at 15w and up, but Atom is still a dog.

• JarredWalton - Wednesday, December 14, 2011 - link

  When you strip out all of the extra stuff in PC Atom (PCI and other legacy interfaces, for example) and add some additional power optimizations, I'm confident that Atom's core architecture can compete with ARM. I don't know that they'll be better with anything in the next year or two, but I certainly think they can compete -- especially if they get in a better GPU than what they have and then manufacture it on their cutting edge 22nm process technology.
  
  FWIW, in terms of pure CPU performance, dual-core Atom N570 on a Chromebook completes the Sunspider 0.9.1 test in 1346ms, which is 25% faster than even the ASUS Eee Prime. Some of that may be Chrome OS, sure, but then Android is basically in the same boat. Cut unnecessary features from the Atom silicon, add in the baseband and wireless support, on-die memory, and a better GPU, and then manufacture that on Intel's 32nm and it should be competitive. Move to 22nm tri-gate and you're looking at potentially class leading hardware. Too bad Atom probably won't get 22nm until late next year (at earliest).

• MonkeyPaw - Wednesday, December 14, 2011 - link

  Hard to say. The Prime, while quad core, runs on the older A9. By the time we see an Intel solution, we might also be dealing with next-gen products like the A15 and Krait running at higher frequencies than Kal-el. While Intel has a lead on process technology, they need to be faster first. They can't afford to have their 22nm products competing against everyone else's 32nm stuff. I guess I figure that mobile CPUs are low-margin products, so they wouldn't pay the bills on the high-end fab equipment.
  
  What is cool at this point with ULV products is the major players are actually a step behind. Tegra2 and 3 are okay, but their GPU and memory bus specs are still behind the curve. To me, that shows how drastic a design change it is for Intel and nVidia to make products that have such low power requirements. The desktop this ain't! :)

• LtGoonRush - Wednesday, December 14, 2011 - link

  I think we've already run up against the fundamental performance limitations that Atom's in-order architecture can deliver. it's not particularly competitive with even a dual-core Cortex A9 given that A9s now exist at comparable clockspeeds, and it's just plain antiquated compared to Cortex A15 (much less AMD's Bobcat-based APUs). The real issue is the lack of a functional GPU in Atom, not only does this mean you can't do much involving graphics (mostly precluding a responsive UI), but video decode performance is crippled as well. That's why Atom-based netbooks ceased to sell, a netbook that can't watch Youtube isn't much of a netbook.
  
  Atom is essentially irrelevent until the Silvermont architecture in 2013, and the single largest reason for this was Intel's inability to foresee the need for a GPU providing even basic hardware acceleration, which is really an unfortunate lack of market awareness.

• MonkeyPaw - Wednesday, December 14, 2011 - link

  I think they see the need, they just think that their almighty x86 architecture can do it all, even though other ways are far better (or at least doing the job already). I bet their Larribee gamble had some part in this, while their lack of real driver support is another matter. Even with Sandy Bridge and Ivy Bridge, I'm yet to be convinced they can actually do graphics. That leaves a lot of things to be reconciled to take ARM down in the near future. Once Intel does figure it out, they will probably apply their "features for money" business model to the mobile space. :(

• dertechie - Wednesday, December 14, 2011 - link

  Here's the thing. For years, Atom has basically been the red-headed stepchild of the Intel lineup. It made crap money, got crap reviews, and was paired with crap chipsets in (mostly) crap designs. Basically, Atom 2006-2010 is what you get when Intel doesn't give a damn, and neither do its partners.
  
  I really don't think x86 is what holds Atom back. That stopped mattering when we hit GHz speeds. Intel's apathy towards the ultramobile sector is to blame there.
  
  Intel's got the R&D muscle to make Atom competitive. I think the change in roadmap priorities (last to 45nm,32nm, but hits 14nm same year as the Haswell shrink) and business reorg is part of shifting some of those resources to make Atom not suck.
  
  Sometime this year, they noticed that superphones and tablets could make lots of money (and replace low end notebooks for pure content consumption). That's when they changed the Atom roadmap. We know Intel executes well. We know Intel knows how to make chips. We know they have a huge foundry tech advantage if they decide to apply it here. I can easily see an Atom Conroe happening.
  
  Or it could just keep sucking. There's no middle road here. Atom will rock in a few years or ARM and Bobcat will bury it.

• name99 - Thursday, December 15, 2011 - link

  "I really don't think x86 is what holds Atom back. "
  
  You don't get it. The problem is not x86 per se (though that certainly doesn't help). It is that keeping all this crap around makes an Atom phenomenally more complex than an Arm. And that in turn means Intel can't rev them as fast as ARM, and can't offer customized versions.
  
  And throwing more money at the problem won't help. When you're dealing with these sorts of insane levels of complexity, you are very much in mythical-man-month territory. Intel undoubtedly hates it --- but design and verification of these things takes forever, and there is no way around that.

• name99 - Thursday, December 15, 2011 - link

  "When you strip out all of the extra stuff in PC Atom (PCI and other legacy interfaces, for example) and add some additional power optimizations"
  
  "Cut unnecessary features from the Atom silicon"
  
  Well some of us have been telling Intel to do that from before there was an Atom. So far, Intel has been absolutely unwilling to ship an Atom that is, apparently, able to boot anything from DOS 1.0 on, and which contains every random piece of crap ever added to the x86 architecture, no matter how irrelevant that might be to the target market.
  
  You can argue all you like about how wonderful some mythic "Atom-lite" might be, but HW developers have to deal with the Atoms Intel is willing to sell today, and the pattern they have developed so far.

• SlyNine - Sunday, December 18, 2011 - link

  I'm not sure how a ARM CPU would compete with Atom, Which I Believe is made to fit a completely different power envelope.
  
  I'm pretty sure even a tegra 3 would not run Windows 7 all that well, if it did support Windows 7. Even under honeycomb Anand said it gets slow if you multitask. Which isn't nearly as bulky as Windows 7.
  
  Atom is not made to run in a smart phones or tablets. Problem is it sucks at what it was made for, even (IF) it's more powerful.

• LtGoonRush - Tuesday, December 20, 2011 - link

  While Windows 7 is relevant for desktops and laptops, the tablet/smartbook/nettop market is going to be a fight between Android 4.0 (Ice Cream Sandwich) and later versions and Windows 8. Both Android and Windows 8 support x86 and ARM, and you should be able to build apps for either OS that are platform-agnostic. This puts ARM in direct competition with Intel's Atom and AMD's Bobcat, a competition that Cortex A9 and later ARM cores are well positioned for.
  
  Intel has tried to push Atom for smartphones, but the same weaknesses that hamstring it on the desktop make it completely unsuitable for smartphones, and no one has ever been stupid enough to build a phone around it.