Intel Atom Z600

[liolio]

I'm not sure that there was anything new in those slides, I decided to post them nonetheless:
http://mobilegeeks.com/leaked-intel-atom-tablet-roadmap-bay-trail-t-valley-view-t-detailed/

Edit
My own sentene sounds weird even to me, I meant in no way that I posted the slides on mobilgeeks or am the source. I'm left wondering about how I should phrase that...

 

[french toast]

I'm not sure that there was anything new in those slides, I decided to post them nonetheless:
http://mobilegeeks.com/leaked-intel-atom-tablet-roadmap-bay-trail-t-valley-view-t-detailed/

Edit
My own sentene sounds weird even to me, I meant in no way that I posted the slides on mobilgeeks or am the source. I'm left wondering about how I should phrase that...

Ha ha

 

[Ailuros]

http://newsroom.intel.com/community...-shot-intel-ships-billionth-communication-soc

Intel has shipped one billion SoCs targeted at ultra low cost, ultra slim phones.

 

[Exophase]

Despite the title this is about cellular chips and not Atom SoCs, isn't it? Did Infineon ship the first 999 million before being bought by Intel?

 

[Laurent06]

Yeah these are Infineon chips, and these contain ARM CPUs and are fabbed at TSMC

 

[Ailuros]

http://www.glbenchmark.com/phonedetails.jsp?D=Intel+Spring+Hill&benchmark=glpro25

Neither springhill nor studybook ring a bell. That 535 should be clocked at 400MHz to yield those results. Anyone have any idea what the device is all about?

 

[Laurent06]

http://www.glbenchmark.com/phonedetails.jsp?D=Intel+Spring+Hill&benchmark=glpro25

Neither springhill nor studybook ring a bell. That 535 should be clocked at 400MHz to yield those results. Anyone have any idea what the device is all about?

http://www.intel.com/content/www/us/en/intel-learning-series/studybook.html

This uses Atom Z650: http://ark.intel.com/products/55664/Intel-Atom-Processor-Z650-512K-Cache-1_20-GHz

That page claims a frequency of 400 MHz.

 

[Deleted member 13524]

First hands-on on a Clovertrail by anandtech:
http://www.anandtech.com/show/6522/the-clover-trail-atom-z2760-review-acers-w510-tested


Compared to Cortex A9, Cortex A15 and apple's swift, the dual-core/4-thread Atom CPU seems to do great in both performance and power consumption.
The problem is with the GPU. The SGX545 seems to suffer quite a bit at 1366*768 and hinders the UI's performance, making the Tegra 3's implementation on Windows RT feel snappier.

 

[dagamer]

First hands-on on a Clovertrail by anandtech:
http://www.anandtech.com/show/6522/the-clover-trail-atom-z2760-review-acers-w510-tested


Compared to Cortex A9, Cortex A15 and apple's swift, the dual-core/4-thread Atom CPU seems to do great in both performance and power consumption.
The problem is with the GPU. The SGX545 seems to suffer quite a bit at 1366*768 and hinders the UI's performance, making the Tegra 3's implementation on Windows RT feel snappier.

I keep wondering why SoC OEMs are using such weak GPUs. TI had the same problem.

 

[Silent_Buddha]

Intel is used to targetting business/corporations. At that level you just need the GPU to be "good enough" with the right check boxes checked. AMD and Nvidia basically took care of everyone else that needed greater graphics performance.

The past couple years they've put increasing focus on GPU's, but it takes a while for a large company (even one as relatively nimble as Intel) to radically change their way of thinking.

Regards,
SB

 

[DavidC]

2 Cortex A9's at 1.66ghz properly optimised and on Intels 32nm HKMG process would destroy Saltwell in every metric including battery life and even die size wouldn't it? (?)

I disagree with the general notion that its nearly entirely due to their process advantage, or even that because before then they didn't try enough, which is why they couldn't break through.

Go back to 2010, when Moorestown was being touted by Intel. There were Android 3.0 Tablet prototypes that performed absolutely HORRIBLE compared to ARM on Android 3.0. They were at 1/3 to 1/4th of the performance in some areas.

Moorestown, despite having similar high-level power management architecture to Medfield were nowhere near close to being viable. The few devices based on it had zero advancement in the power area compared to previous generation. Wait, before you argue that 32nm would have brought lower power: It won't change the fact that Moorestown based devices had no advancement in the idle power area either.

The biggest difference between then and now? Its because Intel started thinking outside of their "box". The first breakthrough is announcement with Google. The second was throwing out the then-current mobile management and getting experienced people like Mike Bell. Those factors have absolutely NOTHING with process, or even architecture. They couldn't break in with Moorestown because their corporate mentality is so ingrained in making PC based devices.

Back to what you were saying. I think Cortex A9 on Intel's process would have slightly lower idle power and little better clocks, but that's about it. It won't be a make-or-break type of deal. Oh, and die size might end up somewhat larger than they do now.

Also consider that Medfield's CPU architecture is minimal change from the Silverthorne one back in 2009. Prescott couldn't survive based on process merits alone, it needed an architecture overhaul. Basically Atom has been "Tick-ing" for years. Which is why Silvermont has that much more potential to be a game changer. It's not only about 22nm, but bringing everything they learned in the past 5-6 years to a new uarch.

 

[rpg.314]

Prescott couldn't survive based on process merits alone, it needed an architecture overhaul.

That's because Prescott needed process merits of the high clock kind. Which weren't to be found around it's time.

Medfield and successors need low power, which Intel can readily provide today.

 

[Laurent06]

That's because Prescott needed process merits of the high clock kind. Which weren't to be found around it's time.

The P4 architecture was inherently power hungry. Deep pipelines and relying too much on replay are bad for power efficiency

Medfield and successors need low power, which Intel can readily provide today.

It's not only a process issue, though obviously one needs a power efficient process which Intel finally has now

 

[Ailuros]

Here's some further digging from Anand for Clovertrail vs. T3: http://www.anandtech.com/show/6529/busting-the-x86-power-myth-indepth-clover-trail-power-analysis

For the time being until I can parallelize performance with power consumption the above results don't tell me much.

 

[Laurent06]

Courtesy of Intel measures, and without using the T3 fifth core. And starting the article by claiming Clover trail beats A9. Yeah sure. Hmm wait Tomshardware has almost the same article.

I'm not saying all of that is BS but it looks so much like an Intel PR push that it stinks

 

[Ailuros]

Courtesy of Intel measures, and without using the T3 fifth core. And starting the article by claiming Clover trail beats A9. Yeah sure. Hmm wait Tomshardware has almost the same article.

I'm not saying all of that is BS but it looks so much like an Intel PR push that it stinks

I've read it with the necessary eyebrow-lifting. However since I personally mostly concentrate on the GPU side of things, I'd still would like to see a full perf/mW comparison and not just some sterile power consumption measurings between the two.

 

[Silent_Buddha]

Here's some further digging from Anand for Clovertrail vs. T3: http://www.anandtech.com/show/6529/busting-the-x86-power-myth-indepth-clover-trail-power-analysis

For the time being until I can parallelize performance with power consumption the above results don't tell me much.

That was interesting. And helps explain why Clovertrail has such a dramatic increase (over doubling) in battery life compared to the original Atom. It would have been interesting to be able to see the same comparison, except with an original Atom slate instead of Tegra 3.

And I guess that's also part of the reason they are so much lighter now. My ~2 year old 11.6" Atom based slate is 2.25 pounds compared to the 1.5 to 1.25 pound Clovertrail slates. The thinness and lightness of current Clovertrail slates was pleasantly surprising to me. I had honestly expected them to be more battery hungry and heavier than they turned out to be.

It's also interesting that part of the power savings compared to Tegra 3, is that Clovertrail is just plain faster. Finishing a task first and going to idle more quickly equates to significant power savings.

It'd be interesting to see how things lined up if Clovertrail had a more performant GPU. I wouldn't expect things to change much, however, as Tegra 3 seems to just use a LOT of GPU power even in tasks that don't require much GPU power.

And I don't see how this could be Intel influenced. Anandtech did spot verify that overall system power consumption was similar to what was being measured with the Intel device. And as Anandtech were free to choose their own applications, it can't be that Intel specifically chose x86 friendly applications.

Regards,
SB

 

[Laurent06]

The fact that T3 power saving fifth core isn't enabled doesn't bother you? The fact that T3 is a year old and supposed to be more power hungry while having less performance than Exynos or A6 doesn't bother you? The fact Windows 8 isn't Windows RT doesn't bother you?

Why didn't Intel show the same kind of comparison on smartphones? They would have the same OS, Android. They would have the choice among more SoC. They could have run games.

 

[Silent_Buddha]

The fact that T3 power saving fifth core isn't enabled doesn't bother you? The fact that T3 is a year old and supposed to be more power hungry while having less performance than Exynos or A6 doesn't bother you? The fact Windows 8 isn't Windows RT doesn't bother you?

Why didn't Intel show the same kind of comparison on smartphones? They would have the same OS, Android. They would have the choice among more SoC. They could have run games.

Is the 5th core really going to help much if it isn't faster than the other 4 cores, thus allowing it to finish tasks faster and get to idle more quickly? A large part of the power savings is just the fact that Clovertrail is finishing tasks far more quickly and thus hitting idle much faster, despite at times being only slightly more power efficient at load.

I don't think the 5th core is going to do much to help. Sure it will help with the purely idle numbers, but otherwise? We're assuming a device is being used, no? This wasn't a test of standby/idle time after all.

And going back to...

And starting the article by claiming Clover trail beats A9. Yeah sure.

Is A9 based Exynos really head and shoulders above Tegra 3 in performance? Honest question, since I don't know. I'm not sure it changes much from a performance standpoint if not.

As to the A6, I know the graphics performance has been significantly boosted while the CPU core has seen some instructions added that give it some features of A15 based cores, so is that a fair assumption of what an A9 based core would do as it isn't truly an A9 anymore, being a bit of a hybrid between A9 and the superior A15.

Also, as mentioned, we'll be able to see how it compares to newer Arm models once the Krait RT slates start to show up. At which point Arm should have a fairly large advantage until 10w IVB shows up. Haswell is when Intel may possibly pass up Arm for good. I'm not sure Silvermont will be able to outperform Arm based offerings in terms of absolute CPU performance as it seems to be focusing more on the GPU. But it should offer compelling battery life.

Regards,
SB

 

[Laurent06]

Is the 5th core really going to help much if it isn't faster than the other 4 cores, thus allowing it to finish tasks faster and get to idle more quickly? A large part of the power savings is just the fact that Clovertrail is finishing tasks far more quickly and thus hitting idle much faster, despite at times being only slightly more power efficient at load.

I don't think the 5th core is going to do much to help. Sure it will help with the purely idle numbers, but otherwise? We're assuming a device is being used, no? This wasn't a test of standby/idle time after all.

You are right. But for instance I'd expect video playback to be light enough on CPU that running the process on the fifth core would be enough

And going back to...



Is A9 based Exynos really head and shoulders above Tegra 3 in performance? Honest question, since I don't know. I'm not sure it changes much from a performance standpoint if not.

http://browser.primatelabs.com/geekbench2/compare/1444425/1445585
Take that with care, but the stream results at the end show the main issue A9 has on T3: memory bandwidth. That's the kind of issue that can slow down heavy apps. Also T3 is still 40nm while latest Exynos are 32nm

As to the A6, I know the graphics performance has been significantly boosted while the CPU core has seen some instructions added that give it some features of A15 based cores, so is that a fair assumption of what an A9 based core would do as it isn't truly an A9 anymore, being a bit of a hybrid between A9 and the superior A15.

Apple A6 CPU, Swift, has nothing to do with A9 or A15, it's Apple own design. It seems to be significantly faster than A9 while power consumption is not too much higher. Of course, comparing it against an Intel chip would obviously be difficult

Also, as mentioned, we'll be able to see how it compares to newer Arm models once the Krait RT slates start to show up.

The problem is that the OS will again be different. But that should be a more fair comparison.

At which point Arm should have a fairly large advantage until 10w IVB shows up. Haswell is when Intel may possibly pass up Arm for good. I'm not sure Silvermont will be able to outperform Arm based offerings in terms of absolute CPU performance as it seems to be focusing more on the GPU. But it should offer compelling battery life.

Haswell ULV 10W is a CPU running at 1.1 GHz (turbo unknown) so it is not obvious it will be as power efficient as existing IVB 17W. I also guess it will be expensive and probably will still need active cooling even in a tablet.

As far as Silvermont goes, it might indeed completely change the ARM vs Intel comparisons. But for it to take over ARM it would have to be much better and not too expensive. The GPU performance is also completely unknown. We'll see how it turns, but the battle will certainly be much more interesting than the current one