Intel's smartphone platforms

Just read Anandtech review/preview, seems like those CPU are little beast, real good performances power consumption is real low too.
I think they have a winner here, the 14nm is coming next year, competition is going to be interesting.
 
Not a bad showing. Great CPU performance at low power. Decent graphics performance, again at low power. Considering this is Intel's first attempt at an ultra low power GPU (previous GPUs in that market segment were all from Imagination right?), that's not bad.

Something that also caught my eye and not directly related to Baytrail. All tablets/notebooks at 599 and above will be Haswell. I'd be interested in seeing a 599 or 699 Haswell tablet.

Regards,
SB
 
It's a mixed bag for me.
CPU looks good as expected.
Consumption looks good too (thanks to the HUGE manufacturing process advantage)
but the GPU is very disappointing. Adreno 320 level for a part coming in 6 months ? T4 and S800 are already faster, A7 should be well ahead too, and real direct competitor Logan will run in circles around it...

edit: now if Merrifield / Morganfield on 14nm comes on time by end of 2014, it will be a totally different story. Like Otellini always said, "you can't beat much better transistors"
 
Last edited by a moderator:
It's a mixed bag for me.
CPU looks good as expected.
Consumption looks good too (thanks to the HUGE manufacturing process advantage)
but the GPU is very disappointing. Adreno 320 level for a part coming in 6 months ? T4 and S800 are already faster, A7 should be well ahead too, and real direct competitor Logan will run in circles around it...

edit: now if Merrifield / Morganfield on 14nm comes on time by end of 2014, it will be a totally different story. Like Otellini always said, "you can't beat much better transistors"

It's pretty significant, IMO, as another stepping stone for Intel.

Some people didn't think Intel would be able to compete on the CPU front in the same power envelope with ARM even with a process node advantage, much less be more performant in the same power envelope.

And for a first generation ultra low power GPU from Intel to be competing with Adreno 320, isn't bad. What is going to be important for Intel is how they iterate on that going forward. Will they bother to try to get a class leading GPU solution? Or will they settle for "good enough" GPU performance?

If this was the Intel from 5 years ago, I'd say they would just settle for "good enough." But after seeing the improvements they made for Haswell's integrated GPU, I'm not so sure if they will be satisfied with "good enough" in the ULP space.

Regards,
SB
 
merrifield/moorefield announced.

http://www.anandtech.com/show/7789/intel-talks-merrifield-moorefield-and-lte-at-mwc-2014

From a gpu perspective, Merrifield is Rogue G6400@533Mhz, Moorefield is G6430@533Mhz.

Intel's off-screen GL2.7 results on reference platforms puts both of them ahead of Apple A7.

That's agressive gpu performance (relative to what Intel has traditionally done on its smartphone socs), and highly competitive.

I will be extremely interested to compare Baytrail graphics performance, with Merrifield/Moorefield :)
 
http://www.anandtech.com/show/7789/intel-talks-merrifield-moorefield-and-lte-at-mwc-2014

From a gpu perspective, Merrifield is Rogue G6400@533Mhz, Moorefield is G6430@533Mhz.

Intel's off-screen GL2.7 results on reference platforms puts both of them ahead of Apple A7.

That's agressive gpu performance (relative to what Intel has traditionally done on its smartphone socs), and highly competitive.

I will be extremely interested to compare Baytrail graphics performance, with Merrifield/Moorefield :)
aggressive for intel yes, but average for the market. In fact they barely compete with last year SoC GPUs. K1 and S805 will destroy Intel effort. Same story again and again...
 
aggressive for intel yes, but average for the market. In fact they barely compete with last year SoC GPUs. K1 and S805 will destroy Intel effort. Same story again and again...

I don't agree (and as I said previously, I'm talking about graphics here).

Moorefield max clock is 533Mhz. A7 in the ipad is around 380Mhz, on what on the face of it, is the same GPU block. That means 2nd half of this year, Intel will launch a 4core soc with 40% more theoretical graphics performance than the A7.

I don't think Intel is shooting to be the graphics performance king, but it will be mid-to-top, and I don't think that any socs out there will "destroy" Intel in the graphics department.

I also don't think you'll see too many K1's in phones at all, nor with the performance metrics equal to what we've seen in the public benchmarks.

Nvidia CEO Jen-Hsun Huang
“Maybe, another way of saying it is what is our non-focus and our non-focus is mainstream phones.”
 
Last edited by a moderator:
It's more like 450MHz for the Apple A7 GPU, but that's besides the point.

As for the usual useless noise in the system we'll see how things will turn our in the longrun as Intel isn't exactly doing well with smartphones so far either.
 
It's more like 450MHz for the Apple A7 GPU, but that's besides the point.
Ok then. I was just basing it on the ipad fill rate, and the fact that IMG generally get close to their theoretical fill rate. Fill rate of around 3000Mt/s on 8 per clock, would be 380Mhz at 100 % efficiency. I suppose 83% isn't too shabby either.
 
Ok then. I was just basing it on the ipad fill rate, and the fact that IMG generally get close to their theoretical fill rate. Fill rate of around 3000Mt/s on 8 per clock, would be 380Mhz at 100 % efficiency. I suppose 83% isn't too shabby either.

That's because you made the mistake and used probably the GLB3.0 fillrate test result which I think uses a healthy proportion of alpha blending (and yes you'll get lower results here as in 2.7 if you check and compare the offscreen GFXbench3.0 alpha blending test between the iPad4 and iPadAir as an example).

Under GLB2.7 you have http://www.glbenchmark.com/device.j...OS&api=gl&D=Apple+iPad+Air&testgroup=lowlevel 3468 MTexel's in the Air's native resolution which gives divided by 8 roughly 434MHz. 450-466MHz IMHO.
 
That's because you made the mistake and used probably the GLB3.0 fillrate test result which I think uses a healthy proportion of alpha blending (and yes you'll get lower results here as in 2.7 if you check and compare the offscreen GFXbench3.0 alpha blending test between the iPad4 and iPadAir as an example).

Under GLB2.7 you have http://www.glbenchmark.com/device.j...OS&api=gl&D=Apple+iPad+Air&testgroup=lowlevel 3468 MTexel's in the Air's native resolution which gives divided by 8 roughly 434MHz. 450-466MHz IMHO.

Indeed I did, clearly all fill rate tests are not equal !
 
Intel was clocking their last generation of SGX XT at 533 MHz, too.

I wonder what kind of power and heat numbers the researchers are seeing at 600+ MHz, even on 28 nm and below processes, and whether Apple would push faster than the 433 - 467 MHz range with a G6630 in an A8 even if it were manufactured on TSMC's 20 nm process.
 
Merrifield/Moorefield headline target segment is supposed to be smartphones. However, even though there has been a paper launch of merrifield some months back, no design wins have been announced.

I noticed today that Asus has announced that their MeMo Pad 8 tablet, will be Moorefield, and they are talking about Aug '14.

That should comfortably exceed iPad air in graphics, and substantially exceed it in multi-threaded CPU performance. Don't know about battery life of course.

Understanding of course that iPad air will be out about a year by then, but likely it will still be the current model in August.

http://liliputing.com/2014/05/asus-memo-pad-8-android-tablet-launches-japan.html
 
It's interesting with these hardware specs (for people like us), but ultimately irrelevant, as mobile platforms aren't performance limited in the same way conventional desktops/laptops are. Mobile software runs fine on portable hardware several years old (which partially explains the runaway success of modern portable devices), so wether this or that SoC offers better performance in a particular way is something that ultimately is not going to offer much concrete benefit.

Battery efficiency on the other hand... :)
 

For these processor, Intel took the PowerVR G6430 GPU and clocked it at 533 MHz, instead of Apple’s 200 MHz.
200 what? :LOL: The frequency difference for the GPU is at 80MHz at best between the A7 and Moorefield G6430.

Something more interesting:

Merrifield
G6400@533MHz (136.5 GFLOPs FP32)
3dmark Ice Storm: 16407

Moorefield
G6430@533MHz (136.5 GFLOPs FP32)
3dmark Ice Storm: 20997

S801
Adreno330@578MHz (148 GFLOPs FP32)
3dmark Ice Storm: 18144

[strike]The interesting part is for the 6400 and 6430 comparison; I think the first lacks FB compression and FP16 SPs, which if true accounts in that particular test for a ~28% difference. [/strike] ***edit: scratch that just further zoomed in to the screenshot and saw the graphics scores of each.

In GFXbench3.0/Manhattan offscreen Moorefield should end up in the 14-15 fps ballpark.
 
Back
Top