Microsoft Surface tablets

[Laurent06]

Android platform and software is mostly build on top of Java (Dalvik VM). Java benchmarks are very much relevant for Android. Only a small fraction of software is build on top of NDK.

Agreed. But we were discussing the CPU performance. Though I agree that in the end what matters is the performance of the CPU + OS/software stack couple.

If you want to benchmark native NDK software, you need to be aware, that the native software needs to be compiled separately for x86 target in order for it to run optimally on ATOM. If you run a native application that only has ARM version, Intel will perform ARM->x86 binary translation for the native machine code. This naturally results in lower performance. Are you sure the Geekbench isn't running on top of binary translation on ATOM, or have they made a separate x86 build for ATOM based Android devices?

Yes, since version 2.3.4 Geekbench is a native x86 app on Android (ref).

BTW I wonder where Houdini (the ARM->x86 translator) performance stands. Obviously for apps that are depending a lot on Android APIs the performance will be OK, but for CPU intensive tasks, I wonder.

 

[wco81]

How well do those Atom SOCs run ARM apps?

Is that Droid RAZR running apps. in emulation? Or is the Atom in that device only good for running benchmarks.

 

[Mintmaster]

Clovertrail is a disaster next to A6x.

Only in GPU performance. The CPU will do fine, as sebbbi outlined. I know JavaScript is highly software dependent, but it's doing fine there.

Don't know about that. They have an architecture license. They don't seem to be using it for consoles.

That's an interesting point, but I think it's just a backup plan. People are spending enough on phones now to get Intel's attention. Intel's process advantage is too large, and customers are disappointed with 20nm offering little power or cost advantage over 28nm at TSMC. By 2014 or 2015, they'll blow ARM offerings out of the water in perf/W.

What exactly would those x86 capabilities be?

Running legacy x86 software?

Most of the mobile software, designed for multitouch, is compiled for ARM.

Legacy x86 is important for business and consumers benefit too, and it doesn't have to be touch optimized if you have a stylus or keyboard/mouse. I know Office runs on RT as well, but Win8 is simply a safer buy. A Clovertrail tablet with a keyboard cover/base is a fully functional notebook, will unquestionably go down in price, and once you have one (or a higher end Ivy Bridge/Haswell based hybrid) it makes $400+ ARM tablets useless.

People had bad experiences with netbooks (yet still bought tens of millions of them) when they were single-core Atoms with 4200 RPM HDDs and poorly optimized browsers and websites not designed for their awful screens. 2GHz dual core Atoms with SSDs showing now tablet-optimized websites in modern browsers is an entirely different story.

 

[Arwin]

You know they did it for app compatibility. But, no, it's not overkill IMO. It's simply beautiful, yet I still can see individual pixel on it when reading text, more so than for an iPhone. (I use it very often laying down, iPad resting in my rib cage, with screen much closer than when sitting upright.) With 16x12 it'd be much worse. Why not go for the best you can do, knowing very well that nobody won't be able to match for a long time, when the opportunity is there?

iPhone 4 and up still has higher PPI than the iPad 3, so that's understandable. But I agree 100%, the screen is so absolutely worth it.

With iOS apps generally written to be fixed resolution, everything would look like crap for existing apps. Slightly higher weight is a decent trade-off.

Absolutely.

Can't wait for the 3096x2048 version, though.

I just read that a 3912x2048 or whatever it is 9.7" screen prototype now exists, with a PPI of 463 or so, being very close to the estimated limits of the human eye ... It's a pretty amazing feat, but we're quite a little way away from getting that in consumer price levels I'll bet.

 

[Arwin]

Incidentally, review is quite promising:

http://www.engadget.com/2012/10/23/microsoft-surface-rt-review/

So, if gaming and music and movies and reading are what you're looking to enjoy, then we might advise sitting this one out for a few months just to make sure that all your bases will indeed be covered. If, however, you're looking for an impeccably engineered tablet upon which you can do some serious work, a device that doesn't look, feel or act like a toy, then you should get yourself a Surface with Windows RT.

 

[wco81]

Legacy x86 software is their ace in the hole. If they only did Windows RT or didn't do a Surface Pro, they'd be starting from scratch.

But they have to hope that desktop x86 software, including Office, is still relevant, if indeed there is a "post-PC" era.

So that is why the Surface tablets are keyboard-centric, really landscape and tabletop-centric. Steve Ballmer was on TV this morning talking about "work and play." Certainly, for Office, they have to hope buyers are looking for tablets they can use at work well as at home.

But outside of Office, what are the heavy "creative" tasks that most white-collar workers engage in and what software would they use? Probably the browser, Office and maybe some vertical applications specific to their industry or company.

So really, it comes down to people wanting Office enough to buy this over iPad or Android tablets. Other popular consumer software (besides games) would be things like Quicken and Turbo Tax.

Probably easier to port Office and personal finance software to ARM than it is to make heavy and expensive Intel devices.

 

[NRP]

I'd be interested in a tablet that can handle Photoshop.

 

[wco81]

More than an Ultrabook?

At a higher cost than an Ultrabook of similar internals?

 

[NRP]

Maybe. It all depends on how well it works. I like to hike Yosemite's back country, so weight and form factor are important. Admittedly a bizarre use case, but the idea is intriguing.

 

[elroy]

I'd be interested in a tablet that can handle Photoshop.

Me too. And definitely more than an ultrabook. I'm actually looking for a convertible - I like the idea of having two devices in one. Something like the Lenovo Yoga at an absolute minimum. Price, and more so weight (when being used as a tablet), are the biggest considerations for me.

 

[rpg.314]

A6x will be slightly faster than Medfield in CPU performance, but I highly doubt it will be anywhere close to Clovertrail.

You can see the comparison between A6 (iPhone 5) and single core Medfield ATOM (RAZR i) here (http://www.anandtech.com/show/6386/samsung-galaxy-note-2-review-t-mobile-/3).

Medfield wins two CPU benchmarks, ties one, and loses one. Medfield has a single ATOM core, and it still can compete very well (win more benchmarks than lose) against the dual core A6. Clovertrail is a dual core version of Medfield. It will be significantly faster than Medfield. A6x on the other hand is only a slightly overclocked A6 (1.3 GHz -> 1.5 GHz if we believe the rumors). I don't see any possibility that A6x ends up being faster in CPU tasks than Clovertrail.

Atom is only in order but 2 way multithreaded. So it gains a lot from 2 threads even with just one core.

IMO, a lot of those wins are based on Atom's clock speed. I doubt that will be good for battery life. Ivy bridge is another matter completely.

 

[rpg.314]

Ha we missed these biased Javascript/browser comparisons so much

What about some benchmark that doesn't rely on the effort put in the browser and the JS engine?

Why are the benchmarks that don't put a premium on browser and JS JIT not biased?

That is not to say that JS is the only thing that matters. It's just one workload, but arguably the most important one for a smartphone.

 

[rpg.314]

By that time we will have quad core Kraits and Swifts ready to launch in devices.

Quad cores are pointless in mobile ATM. Serial performance rules.

 

[rpg.314]

You know they did it for app compatibility. But, no, it's not overkill IMO. It's simply beautiful, yet I still can see individual pixel on it when reading text, more so than for an iPhone. (I use it very often laying down, iPad resting in my rib cage, with screen much closer than when sitting upright.) With 16x12 it'd be much worse. Why not go for the best you can do, knowing very well that nobody won't be able to match for a long time, when the opportunity is there?

Absolutely. It's high time we had high quality screens and great AA by default.

 

[rpg.314]

Xmas above just said that iOS actually uses font AA. Else the way I understand things its actually a combination of 2048*1536 with N samples of font AA. If I didn't understand him wrong that's something that can be beaten easily, since I was thinking all this time that iPad3 doesn't use any font AA at all.

Yes. AA is necessary for font rendering even at 260 ppi since fonts are read slowly so minor jarring is a lot more irritating vs a fast moving game.

 

[zed]

I'd be interested in a tablet that can handle Photoshop.

um they exist already try one of the ultra laptops or so, theyre pretty light nowadays

 

[Ailuros]

Quad cores are pointless in mobile ATM. Serial performance rules.

Let's see what the current Atom successor will look like in a year.

 

[rpg.314]

Only in GPU performance. The CPU will do fine, as sebbbi outlined. I know JavaScript is highly software dependent, but it's doing fine there.

I am sceptical about it doing as well outside JS. It relies too much on a high clock speed rather than architecture which is not a good sign. Silvermont should be a lot better though. If only we had decent mobile benchmarks, sigh...

That's an interesting point, but I think it's just a backup plan.

I had thought about that, but it doesn't seem to make sense. Why would you spend money on it now and then sit on a license. If later on you feel you need a custom ARM core of your own, then you could just negotiate it then. The time taken to design the core would still be the same. Or the backup plan is to make a core now, but release it only when it's deemed necessary. It could be that the plan is to license a custom IP core from one of the architecture licensees but ARM's contracts stipulate they can't do that without an architecture license directly from ARM. But still, why go through the negotiations and the expenditure right now and then sit on the results when it can be done later on as well?

There is something to be said for it being a backup plan. But the specifics of it seem dodgy from a distance.

People are spending enough on phones now to get Intel's attention. Intel's process advantage is too large, and customers are disappointed with 20nm offering little power or cost advantage over 28nm at TSMC. By 2014 or 2015, they'll blow ARM offerings out of the water in perf/W.

Absolutely. Let's hope they bring their A game to Atom the way they did after Netburst and have been doing it since.

 

[zed]

ering little power or cost advantage over 28nm at TSMC. By 2014 or 2015, they'll blow ARM offerings out of the water in perf/W.

too easy inte & performance & blowing away the comp?
Larrabee
he he
the only thing theyll blow is a lot of gas bubbles underwater (but hey todays price of carbon)

 

[Laurent06]

Why are the benchmarks that don't put a premium on browser and JS JIT not biased?

Biased certainly was a poorly chosen word. I should have said misleading for reasons I explained later: JS speed depends too much on browser and JS engine versions.

That is not to say that JS is the only thing that matters. It's just one workload, but arguably the most important one for a smartphone.

I agree, but it's not enough to compare the performance of two SoC.