CherryTrail-T information.

[Nothingness]

Geekbench is crap.

Anyway, the A7 is nowhere near as good as its believed to be. From the 3dmark physics problems (unable to handle random memory access any faster than A6) to Apple's software optimization.

How clever, you have discovered that JS benchmarks depend on the browser! Now redo the experiment on your x86 PC, pick IE9 then Chrome.

The differing Ecosystem makes any comparison difficult to do. Even so past tests have shown that apple can somehow improve performance on its SOC's be almost a factor of two (50%+ performance gain at only 66% the clockspeed) on the same architecture.

Yes, but at the same you dismiss a standalone benchmark without explaining why it is "crap". Ridiculous.

Is the A7 a boss? Maybe, but probably not. IMO, a S800 in an iphone would probably demolish the A7.

You're so wrong it's not even worth discussing. Go re-read Anand review and its analysis of the micro-architectures.

 

[Khato]

4202Y indeed is faster: http://browser.primatelabs.com/geekbench3/compare/207653?baseline=190269

But that's what I'd expect for a 2 GHz Haswell I find the battery results good but not great compared to iPad. Of course that's a x86 and hence it runs Windows for those who need it. Still too expensive for me

Nice find on the geekbench 3 result - I've never had much luck finding what I want from their results browser.

Anyway, point being that Haswell operating in the same ~7.5W thermal envelope demonstrates markedly higher performance than the A7. The fact that it's running at a higher frequency to obtain such doesn't really matter since we have no idea how much A7 power would balloon if pushed to higher frequencies. And yeah, I pay very little attention to the battery life numbers - in this case they're hampered by both the OS and the 13" screen when compared to an iPad. That's why I tend to like the notebookcheck reviews, they've been the only site that even attempts to include power consumption figures. (Which aren't really useful in terms of comparing devices, but they're great for attempting to compare SoCs.)

 

[LogOver]

How clever, you have discovered that JS benchmarks depend on the browser! Now redo the experiment on your x86 PC, pick IE9 then Chrome.

And Geekbench depends on compiler and optimizations. Moreover it has closed source code, so you can't be sure that versions for different platform are actually running the same algorithms/code (you don't get any meaningful results from the benchmark indicating that the calculations are correct).
In addition some results are really ridiculous. For example DGEMM results on Intel hardware is very low. You can get much better (2-3 times better) results by using, for example, open source Atlas library.

For example peak FP performace of i5-4202Y is 51 GFLOPs (using AVX2) or 12.8 GFLOPs (using SSE2). You can get~80-85% of peak perf. in DGEMM using ATLAS. But in geekbench i5-4202Y scores 2.97 Gflops. The same goes pretty much to all geekbench FP calculations.

 

[Enigmoid]

How clever, you have discovered that JS benchmarks depend on the browser! Now redo the experiment on your x86 PC, pick IE9 then Chrome.

Yes, but at the same you dismiss a standalone benchmark without explaining why it is "crap". Ridiculous.

You're so wrong it's not even worth discussing. Go re-read Anand review and its analysis of the micro-architectures.

Yep, and everyone is using Browser benchmarks or geekbench (by Primate Labs which has no special connection to apple at all) so show the A7 in such a powerful light. Who controls apple browsers? Apple of course and with their integration they can optimize to ridiculous levels (how everything is smoother on IOS as it can be customized to the hardware).

Geekbench is crap because its done on different compilers with different optimization flags and often gets much lower results than other programs. Geekbench also places too much emphasis on encryption (which isn't at all the target market for phones).

And to say AT is faultless is wrong. AT remains one of the only sites to consider Haswell a success on the desktop, the R290/ 290X fan business. Anand also never commented on the discrepancy between his iphone 5 launch scores and ios 7 scores in the iphone 5c/s review which should have been big news (in fact there really isn't anything there about IOS7 performance gains).

 

[Homeles]

I find it interesting that Nothingness puts so much faith behind Anand, but is either ignorant of or ignoring the fact that Anand completely disregards Geekbench in regards to cross-OS comparisons.

 

[Shivansps]

Actually BT CPU perf was acceptable, so was the bat time, the IGP was the weak link.

 

[jdubs03]

I think that the performance increase will be significantly higher than 20%, its competitors are likely to be releasing chips that outperform this current generation by quite a bit especially if they are upgrading to 20nm and A57 architecture (plus customization). With the jump to 14nm Intel only has to keep the same power consumption level, allowing for greater increases in CPU performance, and more room for EU's.

I think 50% sounds more reasonable (that is geekbench 3 score 32-bit around 1500, Cinebench single core = .6) They can't sit idly by, and the added arch improvement should allow for larger increases in performance (more and faster memory certainly helps overall snappiness).

I think we'll see devices with similar drain times as the W is likely to be the same or slightly lower.

 

[ancientarcher]

Haha, while I don't mind the Intel-supplied setups I can understand why some would. The impression I got from little tidbits Anand has said (I think mostly on the recent video) was that he was going to dig into it a bit more once he received an iPad Air that he could take apart to instrument.

Either way, it's only going to get more interesting from here. Baytrail is arguably the most efficient low-power processor available while beating or matching its competition. The A7 bests it, but I'd argue that it's straddling the divide between Baytrail and Haswell. (Take a look at notebookcheck's review of the HP Spectre 13 and note that it's i5-4202Y is quite content in a fanless chassis, has power consumption figures that appear to be comparable to or a touch higher than A7, and given that its performance is roughly 75% that of the i5-4200U would still handily beat the A7 in geekbench.) With Cherrytrail we're going to get a marked bump in CPU performance and 2.5x the GPU performance or more... while Broadwell will bring a 30% reduction in power compared to Haswell at the same performance. Intel certainly appears to be setting a pretty decent bar for the competition.

4202Y indeed is faster: http://browser.primatelabs.com/geekbench3/compare/207653?baseline=190269

But that's what I'd expect for a 2 GHz Haswell I find the battery results good but not great compared to iPad. Of course that's a x86 and hence it runs Windows for those who need it. Still too expensive for me

Yes, the i5 4202Y is about a third better than A7. But, you do also realise that the retail price of the Intel Y series processors is ~$304 and Apple pays <$20 for the A7. I don't know why you are even pitting the processors against each other. Unless of course Intel is willing to cut the price of its Y series processors by 95%, there is no danger of world domination by it :whiste:

And of course, the Haswell Y is manufactured in 22nm Trigate and will only see one upgrade in process tech over the next 2 years while the A7 is on a much worse 28nm planar and will see 2 process upgrades in 2 years (first to 20nm planar and then to 14nm FinFet).

So, even assuming higher than average price inflation on 20nm and 14nm for the A7 processor, in 2 years, I am willing to bet that the Apple processor will handily beat the Intel Y series and cost 5-10% of the cost of the Intel processor. who wins

 

[ShintaiDK]

Apple pays <$20 for the A7.

Is that sub 20$ for the A7 including R&D cost? Or is it simply manufactoring cost? And do you have a link for it?

Also dont use Intels list prices as an indication for what OEMs pay. Even endusers at Microcenter dont pay those prices.

 

[Nothingness]

And Geekbench depends on compiler and optimizations.

Like all binary benchmarks. The Geekbench author already has disclosed compilers and flags used.

Moreover it has closed source code, so you can't be sure that versions for different platform are actually running the same algorithms/code (you don't get any meaningful results from the benchmark indicating that the calculations are correct).

What open source benchmark exists? SPEC isn't open source. Have you ever seen its source code? I have.

Also note Geekbench source code is available if you pay a license. http://www.primatelabs.com/geekbench/versions/
In the same way you can get access to SPEC source code (though the price is not the same ).

In addition some results are really ridiculous. For example DGEMM results on Intel hardware is very low. You can get much better (2-3 times better) results by using, for example, open source Atlas library.

For example peak FP performace of i5-4202Y is 51 GFLOPs (using AVX2) or 12.8 GFLOPs (using SSE2). You can get~80-85% of peak perf. in DGEMM using ATLAS. But in geekbench i5-4202Y scores 2.97 Gflops. The same goes pretty much to all geekbench FP calculations.

These are very valid points. But can you prove the same doesn't apply to Geekbench for ARM? Can you also be sure the number of GFLOPS Geekbench reports is comparable to the one other libs report? Also what's the point of comparing a hand tuned library, while at the same time you complain this benchmark result depends on compiler and flags?

Also note this is a Silvermont thread, so AVX isn't applicable.

Note I certainly agree with you that using this DGEMM result as representative of Intel GEMM performance would be very misleading. But using it as a way to compare what various CPU/OS/compilers do on a similar piece of code, it still has some interest.

I find it interesting that Nothingness puts so much faith behind Anand, but is either ignorant of or ignoring the fact that Anand completely disregards Geekbench in regards to cross-OS comparisons.

Where do you see I put much faith behind Anand? I regularly criticize his reviews such as the power comparisons due to them being set up by Intel. But that's not a reason to dismiss everything he writes. It probably escapes you and many other fanboys that it's not because someone likes or dislikes something that he likes/hates everything An example is that I like Intel CPUs a lot, but am not impressed by Silvermont at all.

BTW don't you find it odd that Anand has no issue comparing JS performance using completely different browsers across different OS while at the same time he doesn't want to use Geekbench? Is he sure Intel/MS/Apple/Google/etc. aren't cheating (like the Sunspiders score seem to show)?

As far as Geekbench goes I have *very* good reasons to trust its results even cross OS (though with care), that's why I don't hesitate to use it as a good way to compare CPUs. This does not prevent me from dismissing AES/SHA results. When you'll have spent a few hours playing with Geekbench disassemblies for various platforms and a few more hours running if on simulators, get back to me and we'll start discussing its merits and deficiencies (look at my post history and you'll see I have already warned against comparisons across OS and across compilers without code inspection).

Of course I prefer to use SPEC 2000/2006 and I can assure you I know where Silvermont really stands against the competition with these benchmarks. And Geekbench reflects the SPEC2k differences between ARM and Intel rather well.

 

[Nothingness]

Is that sub 20$ for the A7 including R&D cost? Or is it simply manufactoring cost? And do you have a link for it?

I guess this is the price that comes from isuppli and I agree with you it's very likely to only be manufacturing cost.

Also dont use Intels list prices as an indication for what OEMs pay. Even endusers at Microcenter dont pay those prices.

Definitely correct. And I bet sometimes Intel sells at manufacturing cost

 

[Homeles]

BTW don't you find it odd that Anand has no issue comparing JS performance using completely different browsers across different OS while at the same time he doesn't want to use Geekbench? Is he sure Intel/MS/Apple/Google/etc. aren't cheating (like the Sunspiders score seem to show)?

No, I don't. They and everyone else bench with what they can, and there simply aren't good alternatives.

As far as Geekbench goes I have *very* good reasons to trust its results even cross OS (though with care), that's why I don't hesitate to use it as a good way to compare CPUs. This does not prevent me from dismissing AES/SHA results. When you'll have spent a few hours playing with Geekbench disassemblies for various platforms and a few more hours running if on simulators, get back to me and we'll start discussing its merits and deficiencies (look at my post history and you'll see I have already warned against comparisons across OS and across compilers without code inspection).

All I'm hearing is "I'm right; the experts are wrong." Combine that with your condescending tone, and you'll find out pretty quickly why no one wants to listen to you.

 

[LogOver]

Like all binary benchmarks. The Geekbench author already has disclosed compilers and flags used.

What open source benchmark exists? SPEC isn't open source. Have you ever seen its source code? I have.

Also note Geekbench source code is available if you pay a license. http://www.primatelabs.com/geekbench/versions/
In the same way you can get access to SPEC source code (though the price is not the same ).

He discussed it but proved nothing. To prove fairness of benchmark he just need to open the code... 90% of his "benchmark" are available in open source variants and have much better performance. Comparing geekbench benchmarks to SPEC benchmark is even not funny. SPEC benchmarks are testing very complicated algorithms while geekbench performs basic task or copy available open source benchmarks.
Lets see what he has in geekbench3:
Crypto(SHA, Twofish, AES) - open source code available pretty much everywhere.
BZIP2 - source code and benchmarks: http://www.bzip.org/
PNG/JPG compressor/decompressor - available everywhere. For example:
https://code.google.com/p/jpeg-compressor/
Lua - scripting language. Don't know what script it runs.
Dijkstra - Basic tree search algorithm. Students learn to code it on the first university semester.
Black-Scholes - many open source implementations are available.
Mandelbrot - Simple code. http://en.wikipedia.org/wiki/Mandelbrot_set
Blur/Sharpen Image - see above.
SGEMM/DGEMM/SFFT/DFFT - open source ATLAS/FFTW including benchmarks.
N-Body/Ray trace - many open source implementations are available. I'm sure the author using one of these implementations.
Pretty much every benchmark form geekbench and much, much, much more are available under linux with open source code. There is even application (http://www.phoronix.com/) which combines many open source benchmark in one test suite.

These are very valid points. But can you prove the same doesn't apply to Geekbench for ARM?

It would not apply for ARM, because currently ARM does not support double precision vector instructions.

Can you also be sure the number of GFLOPS Geekbench reports is comparable to the one other libs report?

Yes, I can be sure, because DGEMM is just a matrix multiplication and all efficient algorithms are known for a long time.

Also what's the point of comparing a hand tuned library, while at the same time you complain this benchmark result depends on compiler and flags?

ATLAS/BLAS/LAPACK is standard open source library available for many architectures (including ARM).

Note I certainly agree with you that using this DGEMM result as representative of Intel GEMM performance would be very misleading. But using it as a way to compare what various CPU/OS/compilers do on a similar piece of code, it still has some interest.

In 5 minutes I can write DGEMM implementation (even without ATLAS) which will be at least twice faster on Intel/AMD hardware than his implementation.

Here I found something interesting:
http://support.primatelabs.com/disc...performance-of-4770k-across-linux-and-windows

Author either lying, or have no clue about things he are talking about.

For example:

We certainly agree that the same hardware should perform similarly and we strive for this. We want the scores to represent the hardware performing well, but we also intend the scores to reflect execution performance of real-world application code.

What real world application he is talking about? MAPLE, MATLAB, Mathematica are not real world?
http://math-atlas.sourceforge.net/faq.html#who

Furthermore, if the libraries are proprietary we don't know exactly what optimizations the library performs.

What the f**k he is talking about?
ATLAS uses a BSD-style license, without the advertising clause. ATLAS's license is taken almost verbatim from the example given at opensource.org
http://math-atlas.sourceforge.net/faq.html#license

 

[ancientarcher]

I guess this is the price that comes from isuppli and I agree with you it's very likely to only be manufacturing cost.

Definitely correct. And I bet sometimes Intel sells at manufacturing cost

Agree with both.
However, consider this. Apple spent a ton of money (a technical term, I know) to buy PA Semi and Intrinsity. For them, chip design is a fixed cost, unless of course they want to fire the whole lot of them - which will be insanity.

isuppli quotes $19 for the A7 chip, which is most probably just the manufacturing cost. But, that is the real cost to Apple - the variable cost. Say Apple (or some other OEM) decides to get rid of ARM and go to Intel for the Y series chips. How much do you think they pay post the discount? $20, $40 or $200/$250. I agree there is very likely a discount, but the discount is hardly of the order of 90-95%, which it would have to be for the Y-series chips to be able to compete effectively with the ARM chips (and hence be comparable).

It is no wonder that the HP spectre convertible with the Intel chip is priced at $1,099. Thank the gods Intel doesn't control the low power chip market. That would have meant >$1000 priced tablets and smartphones with Intel charging $500 or more for its chips...

 

[Khato]

It is no wonder that the HP spectre convertible with the Intel chip is priced at $1,099. Thank the gods Intel doesn't control the low power chip market. That would have meant >$1000 priced tablets and smartphones with Intel charging $500 or more for its chips...

It has more to do with the OEMs than Intel. After all, a Dell Venue 11 Pro with the core i3-4020Y and a 128 GB SSD costs $800... guess how much an iPad Air with the same storage size costs? Yup, $800. Feel free to come back with the fact that Apple is charging an outrageous amount for extra flash storage, but that merely proves my point - the manufacturers are happily charging whatever they think they can get away with regardless of actual cost.

And of course, the Haswell Y is manufactured in 22nm Trigate and will only see one upgrade in process tech over the next 2 years while the A7 is on a much worse 28nm planar and will see 2 process upgrades in 2 years (first to 20nm planar and then to 14nm FinFet).

So, even assuming higher than average price inflation on 20nm and 14nm for the A7 processor, in 2 years, I am willing to bet that the Apple processor will handily beat the Intel Y series and cost 5-10% of the cost of the Intel processor. who wins

Good luck with that. The initial tidbits that TSMC teased on their "16nm" process in advance of IEDM certainly look promising - if they can meet those specs it actually will be slightly superior to Intel's current 22nm process. Unfortunately the common platform half of the foundry world hasn't been so forthcoming on their "14nm" process, which typically means that there isn't even a semblance of a process available for presentation. Meanwhile Intel will be introducing Broadwell and Cherrytrail products using an actual 14nm-class process next year and they've already demonstrated a very early Broadwell sample showing the same CPU performance as Haswell while using 30% less power.

Now sure if Apple can somehow obtain the same level of performance improvement, then sure, they'll almost certainly pull ahead of Intel. But that'd require a disruptive, innovative approach to the design as they've already used the majority of the known methods to increase IPC and would be left with tweaking similar to Intel. That still leaves their software advantage of course, but what happens if Intel gets a similar level of optimization?

 

[Ajay]

I guess this is the price that comes from isuppli and I agree with you it's very likely to only be manufacturing cost.

Just as with IOS development, the end users (iPhone, iPad) pay for the development costs - that's pretty much Apple's MO, given its mostly closed ecosystem.

 

[LogOver]

There are additional pieces of information about Cherry Trail
http://vr-zone.com/articles/first-look-new-features-cherry-trails-gpu/64068.html
While the info is mostly promising (especially 16 EU part), some info looks strange. For example one slide stays that Bay Trail does not support Open GL ES, while Intel said exactly the opposite and Android on BT has been already demonstrated and tested
http://www.anandtech.com/show/7314/intel-baytrail-preview-intel-atom-z3770-tested
Also, the author doesn't have a clue about tablet market it seems. He says that there is no design wins for BT.

 

[ShintaiDK]

There are additional pieces of information about Cherry Trail
http://vr-zone.com/articles/first-look-new-features-cherry-trails-gpu/64068.html
While the info is mostly promising (especially 16 EU part), some info looks strange. For example one slide stays that Bay Trail does not support Open GL ES, while Intel said exactly the opposite and Android on BT has been already demonstrated and tested
http://www.anandtech.com/show/7314/intel-baytrail-preview-intel-atom-z3770-tested
Also, the author doesn't have a clue about tablet market it seems. He says that there is no design wins for BT.

I think its due to not being done yet in software. Android devices with BT first comes in Q1 or so. But it is abit of a mishap from the slide creator.

 

[seitur]

What is estimated Quarter in which Cherry Trail-T will release?

 

[jpiniero]

what is estimated quarter in which cherry trail-t will release?

4q 2014

 

[Arachnotronic]

There are additional pieces of information about Cherry Trail
http://vr-zone.com/articles/first-look-new-features-cherry-trails-gpu/64068.html
While the info is mostly promising (especially 16 EU part), some info looks strange. For example one slide stays that Bay Trail does not support Open GL ES, while Intel said exactly the opposite and Android on BT has been already demonstrated and tested
http://www.anandtech.com/show/7314/intel-baytrail-preview-intel-atom-z3770-tested
Also, the author doesn't have a clue about tablet market it seems. He says that there is no design wins for BT.

The slides are quite old, but at least they're something.

 

[sniffin]

An example is that I like Intel CPUs a lot, but am not impressed by Silvermont at all.

Woah there cowboy, that's not how things work around here

4q 2014

Where'd you get q4 from? Seems logical but all I see in the source is 2014

 

[seitur]

4q 2014

thanks. I was hoping for q3 maybe even last month of q2. bummer

 

[jpiniero]

Where'd you get q4 from? Seems logical but all I see in the source is 2014

Just a projection/prediction. There was talk about the 14 nm Atoms being released before the Core products but I am not sure if it is still the case. They may however ship for revenue in 3Q but being able to buy product then is another story.

 

[Homeles]

There was talk about the 14 nm Atoms being released before the Core products but I am not sure if it is still the case.

It's not. The CEO has confirmed that Broadwell is their leading product.