Intel's smartphone platforms

[tangey]

I can't find a thread for medfield, so decided to create a generic intel smartphone thread.

Motorola just announced its first medfield based phone, and they are getting first access to a 2ghz variant. I wonder does it involve an up clock of the sgx540 ?

http://www.techradar.com/news/phone...la-razr-i-announced-with-intel-inside-1098097

 

[Ailuros]

I can't find a thread for medfield, so decided to create a generic intel smartphone thread.

Motorola just announced its first medfield based phone, and they are getting first access to a 2ghz variant. I wonder does it involve an up clock of the sgx540 ?

http://www.techradar.com/news/phone...la-razr-i-announced-with-intel-inside-1098097

http://www.glbenchmark.com/phonedetails.jsp?benchmark=glpro25&D=Motorola+XT890&testgroup=system

Compared to other SGX540 based SoCs I'm not sure whether the frequency is at 400MHz or higher. In any case maybe Kishonti could consider for the benchmarks reading out also GPU frequencies in the future as they do in the majority of cases for CPUs?

 

[tangey]

Anandtech is saying its basically the same chip, CPU defaults to 1.3ghz, but can ramp up to 2ghz instead of 1.6ghz. Given than, it's probably likely the graphics clock remains unchanged.

http://www.anandtech.com/show/6307/motorola-hits-2ghz-with-intel-powered-razr-i

 

[Mintmaster]

Here's some benchmarking vs a dual core 1.5GHz S4:
http://www.engadget.com/2012/09/18/motorolas-razr-i-benchmarks-intel-2ghz-medfield/

 

[Zaphod]

Qualcomm is not impressed. Or maybe they are, seeing how they thought it appropriate to release that ever so slightly slanted video...

 

[Laurent06]

Geekbench on Medfield: http://browser.primatelabs.com/geekbench2/1060511

 

[OlegSH]

Geekbench on Medfield: http://browser.primatelabs.com/geekbench2/1060511

Thanks!
I've forced 1 core for my Transformer Prime overclocked to 1.6 Ghz
adb shell
su
echo "1" > /sys/kernel/debug/tegra_hotplug/max_cpus
("1" - # of active cores)
This works only for rooted devices
Results are here - http://browser.primatelabs.com/geekbench2/compare/1060511/1074377
It's kinda funny that Atom is slower even in comparison with lower clocked single CortexA9 core with single 32 bit memory channel

 

[french toast]

Confirms what most people thought...atom is a dog...even in comparison to lower clocked current/last? Gen smartphone cpus with lower memory bandwidth.

Can't wait for battery tests of that platform..granted 2ghz is probably only turboes at a few second bursts when the thermals allow so...but even standard medfield on advanced mature 32nm hkmg is no award winner in that respect.

Silvermont/next gen atom will reportedly bring much improved ipc, thanks to either full OoO execution or part OoO...and likely allow even greater clock speeds...

Not to mention avx, mature/improved 22nm trigate and likely some kind of rogue set up.
Still that chip will do well to increase ipc by 25%...which will mAke it about on par with last gen ARM!!
It will be going up against beastly cortex a15s clocked at a minimum 1.7ghz..likely more...maybe krait v3 which will be for more powerfull than cortex a9s..not including apples custom designs.

 

[Simon F]

Geekbench on Medfield: http://browser.primatelabs.com/geekbench2/1060511

A bit OT - but why is a primarity test listed in the FPU section? Surely that has to be integer performance? What am I missing here?

 

[Laurent06]

A bit OT - but why is a primarity test listed in the FPU section? Surely that has to be integer performance? What am I missing here?

I don't know how Primatelabs implemented it in Geekbench, but some big number arithmetics is performing better using FPU (look for multiplication by using FFT).

For instance, Prime95 (which is mainly looking for Mersenne primes) uses FPU for all its internal computations.

 

[Ailuros]

OT but PURE went into the digital TV recorder business with an Intel SoC: http://www.pure.com/tv/

http://www.pocket-lint.com/news/47647/pure-avalon-300r-pictures-and-hands-on

 

[Simon F]

I don't know how Primatelabs implemented it in Geekbench, but some big number arithmetics is performing better using FPU (look for multiplication by using FFT).

For instance, Prime95 (which is mainly looking for Mersenne primes) uses FPU for all its internal computations.

I thought the FFT approach only becomes cost effective for massively huge numbers which I guess you'd get with Mersenne primes. For smaller, more pratical, ranges such as those of use for RSA etc, aren't (stretching my memory here) Karatsuba-Ofmann or Toom-Cook more efficient? Those will be integer-based (IIRC).


[update]Hmmm... the "FFT approach" seems to work in integer as well.

 

[Laurent06]

I thought the FFT approach only becomes cost effective for massively huge numbers which I guess you'd get with Mersenne primes. For smaller, more pratical, ranges such as those of use for RSA etc, aren't (stretching my memory here) Karatsuba-Ofmann or Toom-Cook more efficient? Those will be integer-based (IIRC).


[update]Hmmm... the "FFT approach" seems to work in integer as well.

Integer approaches are not necessarily faster and the threshold to switch from "standard" to FFT depends on many parameters. You'd be surprised how "small" that threshold can be. See gmp mul FFT threshold; note GMP uses integer FFT.

Anyway all of this is moot, because how a benchmark does something isn't always the best way it should be done.

The Geekbench doc says this:

Primality Test performs the first few iterations of the Lucas-Lehmer test on a particular Mersenne number to determine whether or not it is prime.

So it could be some rather big exponent, given that only a few iterations are done.

 

[Laurent06]

I dug further and by looking at strings I found in geekbench, it looks like the prime test is using lucdwt.c which definitely relies on FP numbers for its arithmetic operations.

 

[Simon F]

I dug further and by looking at strings I found in geekbench, it looks like the prime test is using lucdwt.c which definitely relies on FP numbers for its arithmetic operations.

This? Glad they avoided RSI by skipping comments

 

[french toast]

Well, looks like Intel has pulled out a winner with medfield..
The moto razor i...at least in this review..pulls out very fast performance thanks to 2ghz atom..
We expected that due to the turbo clock speed...what I didn't expect was the chip to match or exceed the snapdragon s4 equipped razor M in performance...and beat it in battery life..incredible when you consider the snapdragon is on 28nm and lower clocked!

For those of you that will say medfield has crap ipc and relelively poor app selection...including adobe flash/air/games...fair enough..but it's looks like a very very competitive SOC from where I am.

http://www.engadget.com/2012/10/04/motorola-razr-i-review/

 

[Laurent06]

Except in SunSpider it is slower than Razr M. So what? You read it wrong?

I stand to my belief that if the price of smatphones using Medfield wasn't so low it would be a completely uninteresting chip.

 

[Silent_Buddha]

That is hugely impressive, even considering the relatively lackluster sunspider showing. Now to see what Intel can do once they start to leverage their process advantage.

Regards,
SB

 

[tangey]

A sighting of merrifield ?

http://www.slashgear.com/intel-redh...show-dual-core-merrifield-processor-29262496/

 

[Laurent06]

I guess it's Z2580, not Merrifield. The latter is supposed to come with Intel IGP, while thhe former is coming with SGX 544MP2.