Intel Shows Off Competitive Medfield x86 Android Power & Performance
by Anand Lal Shimpi on December 21, 2011 4:09 PM ESTYou may have seen some Medfield related news today - I thought I'd chime in with some of my thoughts on the topic. As a recap, in 2008 Intel introduced its Atom microprocessor - originally designed for MIDs (Mobile Internet Devices) and repurposed for use in netbooks. In 2010 we got a more integrated version of Atom called Moorestown designed for smartphones and tablets. Intel had basically no success with Moorestown, although Pineview (its netbook counterpart) was still used in netbooks. In our original architecture piece I called Moorestown the "two-chip solution that uses five chips". Despite physical real estate requirements, Moorestown could fit into something that was a bit larger than an iPhone at the time. At the time everyone assumed Moorestown didn't make it into a phone because of unreasonably high power draw, but no one was able to back up those claims with actual test data. Intel will tell you that Moorestown never made it anywhere because it lacked platform focus. Intel was off chasing Moblin/MeeGo and Android wasn't a priority. Things have obviously changed since then.
Medfield is the 32nm successor to Moorestown. Intel hasn't disclosed many details about Medfield's architecture, but we do know that it is a more integrated single chip design (compared to Moorestown's "two-chip" solution). We also saw Intel's Medfield reference smartphone and tablet, both running Android, at IDF. While this is still far away from announcing an actual design win, it's clear that Intel is finally making progress in the right direction. Couple these recent milestones with more recent restructuring inside Intel and it looks like the company is finally preparing to really enter the mobile market.
This brings us back to today's news. The information disclosed today came from an investor conference earlier this month. Above is a slide from the aforementioned meeting.
Intel has carefully removed the names of the devices in this chart. Barring any outright deception however, there seems to be potential in Medfield. One would assume that device manufacturers are given access to this (and more) performance data. One might also be able to infer that if Intel does indeed have a Medfield design win (hopefully more than one), then these numbers might be fairly convincing. The war begins in 2012...
Source: Intel [pdf]
Facebook
Twitter
RSS
39 Comments
View All Comments
Visual - Friday, December 23, 2011 - link
I don't care for a x86 android phone. Not unless Wine runs on it anyway.I am however curious about a x86 Windows device in a similar form factor. Maybe it is time for the UMPC comeback.
But aren't we forgetting something before we can even talk phone-sized form factor?
I am still waiting to have a decent variant in the comparatively much larger (and so supposedly easier to get) form factor of 9-12 inch tablets. We do have a few options, but they fail in comparison to AMD Bobcat alternatives, and even those aren't too great.
And last, it's not just the CPU that needs focus for a "design win". When will some manufacturer finally comprehend that they can not cut corners with crappy TN panels with resolution worse than my wrist watch, passive touchscreens with no stylus, or only the minimum integrated GPU even for when their device is plugged in or docked, etc?
Exophase - Friday, December 23, 2011 - link
Regarding the comment about how Moorestown was dismissed for being too power hungry w/o evidence, there is at least one data point: Fujitsu LOOX F-07C. This phone combined both a Z600 processor at 1.2GHz meant for running Windows 7 and an unknown ARM SoC meant for running Symbian.Here are some reviews:
http://www.engadget.com/2011/08/26/fujitsu-loox-f-...
http://www.pocketables.net/2011/10/fujitsu-loox-f-...
The phone is equipped with a 1400mAh battery, which is pretty standard for smartphones today. Windows 7 mode runs for only about 1-2 hours. This was despite the Z600 being clocked at a meager 600MHz, which would put its typical performance far below what a 1+GHz Cortex-A9 would give you. And the unit got hot just by downloading things. Part of this is Windows' fault, but that only extends so far.
ssj4Gogeta - Monday, December 26, 2011 - link
Was the Z600 made on an LP process?Exophase - Tuesday, December 27, 2011 - link
Allegedly, but bear in mind that this has more to do with idle power consumption than active.LtGoonRush - Tuesday, December 27, 2011 - link
VR-Zone posted some Medfield benchmarks and it looks pretty damning, only a 40% performance increase over a Tegra 2 in CaffeieneMark 3 with a 1.6Ghz Medfield. I don't see how it has any chance of standing up against Tegra 3, much less anything based on Cortex A15 or Qualcomm Krait processors.Link: http://vr-zone.com/articles/intel-s-32nm-medfield-...
ChrisOz - Wednesday, December 28, 2011 - link
If I remember correctly x86 (Intel) was able to beat the pure RISC chips through being ahead on the process curve. This allowed them to offset the disadvantage of CISC instruction set decoding with more transistors for better RISC cores, faster switching ie higher frequencies and cheaper chips.I am suspicious that Intel may not have the head room to be far enough in front to offset x86's problems. I remember reading somewhere that there are only 2-3 further design rule shrinks (16 nm, 10/12nm and may be 8 nm) before chip makers have to look else where for performance improvements. This doesn't give Intel much time to get in front particularly with ARM manufactures already jumping to 28 nm which will put them ahead of Medfield on 22nm both in power and performance.
ARM is cheap, doesn't have to worry about x86 overhead and the software is legacy ARM in the mobile space so x86 loses unless it is much better. It will be interesting to see what Intel can do.
skydrome1 - Wednesday, December 28, 2011 - link
I really think that they have an uphill challenge. Tegra 2 isn't Medfield's competition. Tegra 3 is. Qualcomm's Krait is.I can only hope that Intel manages to push Silvermont to Q4 2012, or it might be quite problematic for them. After all, by then, A15s will be in the market too.
Since Silvermont is speculated to be OoO and have a new architecture, it has hope of really beating (or perhaps tying with) ARM.
Another thing that concerns me is the GPU. What GPU is Intel running? I heard that previous Atoms use the SGX535. Is that right? Hopefully Medfield has an SGX 544MP2 (not 543, Intel would need DX support).
Last chance, Intel. Let's see how you dance.
slugrail - Monday, January 2, 2012 - link
I don't think SGX544 (or at least the dual-core configurations) will be out until NEXT year. And most likely the more valuable choice would be to cut off and use a SGX543MP4 instead (delivers better performance overall than the 543MP2 - if I'm not wrong)extide - Friday, December 30, 2011 - link
This battle seems to be more interesting to me than the desktop x86 battle. Again we see Intel coming from behind but with a TON of money and world class fab capability. 2012 and 2013 will be fun to watch, for sure.