Intel's smartphone platforms

Very interesting move, considering that not all Android Apps can run on x86. This is either Samsung's #1 or #2 tablet, so it's a significant risk.

My theory? Samsung wants a little experience making a high volume, low cost x86 tablet so that they get Silvermont right. Presumably they'll be able to share a lot of design work and parts between the Android and Win8 versions.

It's likelier that Samsung has an internal problem producing enough of its own whatever SoCs to cover all markets they want to adress. Note that they haven't picked the highest end Clovertrail+ from what I can see.
 
Very interesting move, considering that not all Android Apps can run on x86. This is either Samsung's #1 or #2 tablet, so it's a significant risk.
I don't know why they made that move, but the Tab 3 7" will be using a Marvell chip and will probably sell much more.

My theory? Samsung wants a little experience making a high volume, low cost x86 tablet so that they get Silvermont right. Presumably they'll be able to share a lot of design work and parts between the Android and Win8 versions.
It sounds plausible they are indeed getting experience building Intel based Android tablets. Because for Windows tablets they have a long experience ;)
 
It's likelier that Samsung has an internal problem producing enough of its own whatever SoCs to cover all markets they want to adress. Note that they haven't picked the highest end Clovertrail+ from what I can see.
That may be true, but it doesn't explain why they used x86 as opposed to Qualcomm or TI or NVidia other ARM SoC.

I don't know why they made that move, but the Tab 3 7" will be using a Marvell chip and will probably sell much more.
Wow, really? So even within the Galaxy Tab 3 line they're mixing x86 and ARM?

Because for Windows tablets they have a long experience ;)
Not cheaper ones. If they can make a $250 Android tablet with an x86 SoC, then they can use almost the same design with Win8.
 
Last edited by a moderator:
That may be true, but it doesn't explain why they used

Because they don't have enough of their own SoCs to cover any relevant demand maybe? If you meant why they used Intel Clovertrail+, it's a nice mainstream solution compared to GalaxyS4 and they wouldn't want to give Qualcomm exclusivity in the end now would they?
 
Marvell must be giving Samsung a very good deal on PXA986, that or Samsung Mobile really does want to test the waters with everything they can. This SoC looks positively dated and I'd just about written off Marvell as being a viable competitors in the western mobile market (I know they're big in China.. someone please correct me if they actually do sell a lot in Samsung's bigger markets, but I've always had a hard time finding products that used them). I'd actually forgotten they were now licensing Cortex-A9s and Cortex-A7s. I'm astonished that even they don't find their own cores suitable even though they're supposed to be hitting similar targets.

I wonder if Marvell has even moved past 40nm TSMC for this. If not could this really look good in any way vs 32nm Exynos 4? Including the dual cores if necessary.
 
I saw 45nm for the PXA986..

The only advantage I see before the 32nm Exynos 4212 is the integrated baseband..
 
ASUS announced the Fonepad Note, and it uses CloverTrail+.

I'm really surprised that Intel is getting design wins in Android. They must be getting either very good perf/power or pricing for them to deal with the customer service headache of not being able to run every Android app.

I'm not making too much of a deal out of this, am I? Devs do need to recompile apps for x86 unless they're pure Java, right?
 
I assume some (software) binary translation will occur, but I wonder if this will be "good enough" for performance intensive applications (e.g. games)?
 
Mintmaster: your questions make me wonder if you had the chance to read several articles on Medfield and Clovertrail published in the last year or so measuring performance, power & compatibility with Android apps.

I am surprised by you being surprised :)
 
I'm not making too much of a deal out of this, am I? Devs do need to recompile apps for x86 unless they're pure Java, right?

Android NDK uses a fat binary format that includes different compilations, devs can choose to target x86 as well as different ARM instruction mixes with no extra effort so long as it compiles. But this doesn't apply to older software unless it has been updated. And developers can still choose not to do it for whatever reasons.

Intel does have a translator too, but unless they've improved it a lot since they published their paper on it the performance will be < 50% native when running translated code (so not including time spent in system libraries, syscalls, etc). The code that needs translation the most - stuff using NEON intrinsics for instance, that may lack a full x86 equivalent - may also be hit hardest by translation overhead.

nAo said:
Mintmaster: your questions make me wonder if you had the chance to read several articles on Medfield and Clovertrail published in the last year or so measuring performance, power & compatibility with Android apps.

I am surprised by you being surprised :smile:

I haven't seen anyone provide a comprehensive overview of Android game compatibility and performance on Medfield or Clovertrail+. The closest I've seen is a couple of people who have said something along the lines of "we tried a bunch of applications and almost all of them worked." I did see a couple reports of games which didn't run well, but it was hard to tell if ARM translation had anything to do with it or if it was purely down to Medfield's weak GPU.
 
I haven't seen anyone provide a comprehensive overview of Android game compatibility and performance on Medfield or Clovertrail+. The closest I've seen is a couple of people who have said something along the lines of "we tried a bunch of applications and almost all of them worked." I did see a couple reports of games which didn't run well, but it was hard to tell if ARM translation had anything to do with it or if it was purely down to Medfield's weak GPU.
I also wonder if the Google App store doesn't simply hide apps that wouldn't work or would work badly on Android x86 due to translation.
 
I see intel has teased a bit of info on merrifield at computex.

Kilroy promised a 300 percent improvement in graphics and a 50 percent improvement in overall performance compared to predecessor chips, as well as "support for our high resolution 2560x1440 display".
......
Kilroy would not divulge any more about Merrifield "until Barcelona next year", referring to Mobile World Congress in February.
"Barcelona 2014, you can look for that phone to be shipped," Kilroy said.

Well its impossible to know if "predecessor chips" are medfield or clovertrail+. If CT+, then I'd suggest they would require rogue to get x3, unless they are looking at 544mp4@750Mhz. I assume CT+ is already around x3 medfield in graphics terms.

http://www.v3.co.uk/v3-uk/news/2272546/intel-merrifield-chip-arrives-on-first-smartphone-at-computex
 
I assume CT+ is already around x3 medfield in graphics terms.

Samsung GT-P5210 (Clovertrail+)
GLB2.7 = 392 frames 7.0 fps
GLB2.5 = 2841 frames 25.1 fps

Asus ME371MG (Medfield)
GLB2.7 = 98 frames 1.8 fps (3.9x times)
GLB2.5 = 948 frames 8.4 fps (3.0x times)

You know hypothetically (warning: strict speculative math) 21.0 fps or else 1176 frames in GLB2.7 doesn't sound like a whole damn lot. If that's a smartphone platform yes it's most likely Rogue IMHO, but nothing all too fancy either. Rather the predictable *yawn* level compared to the competition's high end.
 
Mintmaster: your questions make me wonder if you had the chance to read several articles on Medfield and Clovertrail published in the last year or so measuring performance, power & compatibility with Android apps.
Oh, I know it's a very solid SoC from a perf & power standpoint. However, compatibility seems to be an issue for some notable apps.

I've only done a cursory googling, but it's pretty complicated to get Flash working, Firefox is still in the test phase, and various apps don't work. These probably aren't dealbreakers for most people, but it I'm pretty sure companies are leery about CS calls regarding app compatibility (or simply being unavailable).

Exophase sums up the reasons for my surprise.
 
Really? That looks low compared to their previous claims and would place Merrifield at a similar position against existing A9 chips: http://browser.primatelabs.com/geekbench2/compare/1970892/2010441

1426x1.5 ~= 2139 for the K900 against 2128 for the Note II.

I guess the 50% is wrong or Merrifield will lag behind existing chips in raw CPU performance.

Did I miss something?

Seems to suggest that Merrifield, or at least the one they're talking about, will be dual core. 50% improvement could be an average over lightly threaded (<= 2 threads) performance, where it could perform better than that if coming in at a higher peak clock, and highly threaded performance where it could come in worse due to lack of HT.

So yes, a quad core A9 would probably beat that in highly parallel tests, for whatever that's actually worth on phones.
 
Seems to suggest that Merrifield, or at least the one they're talking about, will be dual core. 50% improvement could be an average over lightly threaded (<= 2 threads) performance, where it could perform better than that if coming in at a higher peak clock, and highly threaded performance where it could come in worse due to lack of HT.

So yes, a quad core A9 would probably beat that in highly parallel tests, for whatever that's actually worth on phones.

I didn't listen to the presentation, but I'm assuming he is being misquoted in the article, or the author was mixing up baytrail and merrifield.

"Flashing an "early sample" of a Merrifield-powered smartphone during a keynote in Taipei's International Convention Centre, Intel VP of product marketing Tom Kilroy said that the Merrifield 22nm Atom SoC has a quad-core CPU."
 
I didn't listen to the presentation, but I'm assuming he is being misquoted in the article, or the author was mixing up baytrail and merrifield.

"Flashing an "early sample" of a Merrifield-powered smartphone during a keynote in Taipei's International Convention Centre, Intel VP of product marketing Tom Kilroy said that the Merrifield 22nm Atom SoC has a quad-core CPU."
I can't say if he was misquoted, but I found that +50% reference in many places all over the web (which I admit doesn't prove it's correct).

During Silvermont presentation, Intel claimed (cf. slide 34 of the presentation) :
- 2x over Z2580 for 1 thread vs 1 thread
- 2.8x over Z2580 for 4C4T vs 2C4T.

So that 1.5x looks really low if indeed Merrifield is a 4-core chip. The alternative explanation is that Intel claims 1.5x over all smartphone chips, which I would seriously doubt.
 
having now listened the presentation, the relevant section of it he is talking about both baytrail and merrifield, but its clear to me that that quad-core remark related to baytrail.
 
Back
Top