Intel Atom Z600

I would assume (and have been told) that Intel's processes are far denser than competitors, including Samsung.

So if A5X's 543MP4 is 80mm^2 on Samsung's 45nm, it's not inconceivable (yes, I know what that word means) that it can be 20mm^2 on Intel's 22nm.

 

6x the performance of Cedar view is a crazy graphics number this is where that manufacturing advantage starts to show and perhaps where many overlooked..most have been concentrating on X86 effeciency/performance...Graphics, memory, IO, video enc/dec, and baseband will all benefit from the advanced process....very worrying for Qualcomm and the likes..

 

Comparing to licensees like TI for an indication, Apple certainly doesn't have SGX optimized for area with their fab partner. While they have the luxury of trading die for better power/heat characteristics. Intel, fully integrated along the production chain, has that luxury too and also apparently hasn't been optimizing its SGX implementations for area.

I just don't think Intel will be using die area as their primary criterion when deciding between the IP of PowerVR or their internal group.

 

Medfield numbers are up in the Lenovo K800...comparing to Tegra 3/Snapdragon S4/Exynos 4210..
http://www.engadget.com/2012/04/06/lenovo-k800-benchmark/#disqus_thread

Note, that im not sure software is equal....If Charlie is right about Silverthorn only being partially OoO = 25% IPC increase over Saltwell.... along with early reviews on Ivybridge saying it runs a bit hotter than Sandybridge...then Intel may not have it all its own way next year...

 

The results look odd: superlinear scaling for Linpack on One S (dual core), about 3.1 for One X (quad core) and 1.5 for Note (dual core). That doesn't make sense.

Also the Lenovo is still using Android 2.3.7. For an upcoming device that looks strange.

All in all, I don't think this brings more concrete information than what Intel already said months ago. Still waiting for real reviews and comparisons.

 

The Linpack that they're running isn't very well written and isn't really compute limited. It's actually mostly ld/st bound provided the CPU can do serial VDIV's fast enough...

Even real reviews and comparisons don't usually tell much. Mobile benchmarking is...not in a good state.

 

I agree this Linpack is poor but it doesn't explain the superlinear speedup on the S4

I certainly agree. But it will be better than getting figures from Intel or from unknown sources.

 

A multithreaded program that isn't compute limited can scale superlinearly with a shared cache.

I thought the benchmarks were run by the reviewer using the test model.

 

You mean the threads share data

My understanding is that Intel prohibited public benchmarks until release of the phones, but I might be wrong.

 

THe indian version of intel's medfield reference handset (XOLO Lava 900) is showing up on glbenchmark.

http://www.glbenchmark.com/compare....LO Lava X900&D2=Archos 80G9&D3=Google Nexus S

25% lower figures compared to the Vizio VTAB3010 that was posted a while back (which didn't have any low level figures posted). 27 for Egypt Versus 36.

The stand out thing for me in the above comparsion with nexus S and Archos 10G9, is the fill rate. x7 the fill rate of the nexus S and x3 the fill rate of the archos. Most of the triangle tests are x5 the nexus S too.

 

It seems that Intel is finally "there"
http://www.anandtech.com/show/5770/lava-xolo-x900-review-the-first-intel-medfield-phone/1

 

So it's just middle of the pack in battery life and behind current generation of ARM chips performance wise. Add to that there's no guaranty that the native apps you might be using will work. Intel might be "there", but I'm unimpressed (of course I have a bias ).

 

Not really high end, but a good phone, I would not discard it personally 9and so as a costumer) on X86 / perfs alone.
In the mid range I would consider it on the whole package (form factor and look, camera, android version, etc.) and price.

I'm pretty impressed my-self one core with SMT keeps up with some dual or quad cores. Sometime it beats them.
GPU is nothing to wow at but it doesn't seem to be the goal of the design.

I'm willing to see how their high end solution will fare in high end smart phones and tablets.
Intel may let a bit of GPU perfs out of the design but for the devices main usages it's a bit irrelevant.

They have something that to me looks tiny, cheap and performant, it's barely bigger than a tegra2 and beat the crap out of it.

Taking in account how they've been mocked and how they are now I think it's an achievement.

Sadly I'm close to think "the marking is on the wall", Intel is "there" so now should start the lithography war and well...

As a consumer I pray for competition trust me. But when I read the concern of some analysts wrt to Intel (not here more financial analysts) I always though damned not a mystery why economy is in the wall if those people are that clueless. Intel should never be taken lightly.

 

Which in itself is quite an endorsement of the platform. Many have until recently said that Intel would never ever get a workable platform, x86 would always be much more power hungry than ARM, not integrated enough etc. That their first significant x86 handset can work out more or less average across the board is impressive.

When they bring their process advantage to the atom smartphone platform, be interesting to see if they actually become the best solution in terms of power/performance, which is their best bet of being relevant in the sector. x86 compatibilty on its own has little relevance, unless windows becomes a really big thing on smartphones AND WOA doesn't pan out.

 

On single-threaded micro-benchmarks that seem to correlate directly with clockspeed even across more parallel architectures.

Agreed, and it may not even be all that important for the majority of use-cases. But for some reason, a lot of people care about it

I'm more concerned about integer and FP performance for OS and heavy app usage cases. But mobile benchmarking is not in a very good state to tell such things.

 

At least people have more than Sunspider and Browsermark this time around - many thought Sunspider's performance lead was indicative of performance across the board.

Unfortunately, Intel's massive Javascript advantage is still skewing the Browsermark and Vellamo scores. Would be interesting to see what the breakdown of the Basemark scores are, because it's not all CPU bound.

And as we all know, the Linpack bench also has precious little to do with actual vector FP capabilities.

Well if it wasn't any good it wouldn't have been their first significant handset. It has to at least have some minimum of quality to make it into products that anyone wants to sell. Moorestown made it into a phone or two in limited distribution and it was awful. If it wasn't awful it probably would have been more significant too.

I've never been among the "x86 is intrinsically bad for mobile" camp. I'm more of the opinion that Atom in its current incarnation leaves a lot to be desired. But with the state of benchmarking for phones today being a circus show and battery tests being about nothing but web browsing I'm not sure people care about pushing performance in phones as much as they think they do.

 

As it currently stands, the Linpack on Android doesn't even indicate scalar FP performance.

 

I wish someone would tell Anand that. He's been cluelessly repeating this nonsense for years now.

 

The high-end Medfield variant, the Z2580 with its 544MP2, appears to be getting an early test.

http://www.glbenchmark.com/phonedetails.jsp?D=Intel+ctp_pr1&benchmark=glpro21

Judging by the relatively small increase in pixel fill from the existing Z2460 score, clocks and software may still have a ways to go. What is interesting is that the CPU measured at 2.0 GHz, which is above even the Z2580's turbo high of 1.8 GHz. Intel had raised that cap on the Z2460 before it, so maybe they're trying that again.

 

if it allows running any linux distro, even those that only support x86, it could be funny. might require virtualization software, in particular you're not guaranteed to find a BIOS or UEFI and a PCI bus, and I wonder about intentionally crippled features (64bit, virtualisation instructions)

with USB host and display output you would plug it into full keyboard and monitor, and run a XP or windows 7 VM if it's what you want. so, a phone that is not quite a PC but you could use it as if it were a PC