Apple A6 SoC outperforms Intel Atom in JavaScript test(appleinsider.com)
appleinsider.com
Apple A6 SoC outperforms Intel Atom in JavaScript test
http://appleinsider.com/articles/12/09/19/apple_a6_soc_outperforms_intel_atom_in_javascript_test
6 comments
Very interesting and I look forward to a more detailed analysis of the new cpu in the iP5.
I'd be interested in how they handle memory managment/interface and suspect they have done alot of work on that front which would benifit graphics as well as general processing. Might of added a large cache onchip and that in itself will make alot of benchmarks whcih dont have much else running go alot faster than you would expect. This may be the case, we just don't know.
Either way you look at it this has to be unique as far as apple phone releases go with lots of people expecting lots and then still falling in love with what is released and with this new one the general press are starting to lose the glaze in there eye's and in parts poo pooing the new iPhone, which is funny as this does actualy appear to be a good product release with perhaps alot of understated features which do more than back up the claim of 2x faster. This in a world were people are expecting any claim of x amount faster to be backed up with small print outlining the conditions that claim is true. but from what it appears as a whole this 2x claim is a very solid claim in many area's of there new phone.
I'm not so sure apple have as some are indicating redesigned a whole ARM core, but they certainly could of very easily replaced parts of it and expanded other area's and in that adding cache is one area were mobile phone chips will happily accept and get a good return. Also when you write the OS and have a chip like ARM which you haeve more control to add and change things then optimizing for your operating system by adding a few extra instructions of tweaking the area's your profiling tools have identified as bottle necks and you start to get something realy nice.
I'm more a Android chap and never owned a Apple product but this does not stop me appreceiting there finer works of art and for me the iPhone5 has that gut feeling of being a good solid design. OK so it has no NFC but Apple don't do there own credit card, do they, currently - watch this space.
I noticied a few comments saying the A6 is dual and this intel is single core, well there be quad cores and they still aint betean it, be looking forward to some more hands on tests with details - especialy the memory interface and amount of cache memory on that chip.
I'd be interested in how they handle memory managment/interface and suspect they have done alot of work on that front which would benifit graphics as well as general processing. Might of added a large cache onchip and that in itself will make alot of benchmarks whcih dont have much else running go alot faster than you would expect. This may be the case, we just don't know.
Either way you look at it this has to be unique as far as apple phone releases go with lots of people expecting lots and then still falling in love with what is released and with this new one the general press are starting to lose the glaze in there eye's and in parts poo pooing the new iPhone, which is funny as this does actualy appear to be a good product release with perhaps alot of understated features which do more than back up the claim of 2x faster. This in a world were people are expecting any claim of x amount faster to be backed up with small print outlining the conditions that claim is true. but from what it appears as a whole this 2x claim is a very solid claim in many area's of there new phone.
I'm not so sure apple have as some are indicating redesigned a whole ARM core, but they certainly could of very easily replaced parts of it and expanded other area's and in that adding cache is one area were mobile phone chips will happily accept and get a good return. Also when you write the OS and have a chip like ARM which you haeve more control to add and change things then optimizing for your operating system by adding a few extra instructions of tweaking the area's your profiling tools have identified as bottle necks and you start to get something realy nice.
I'm more a Android chap and never owned a Apple product but this does not stop me appreceiting there finer works of art and for me the iPhone5 has that gut feeling of being a good solid design. OK so it has no NFC but Apple don't do there own credit card, do they, currently - watch this space.
I noticied a few comments saying the A6 is dual and this intel is single core, well there be quad cores and they still aint betean it, be looking forward to some more hands on tests with details - especialy the memory interface and amount of cache memory on that chip.
Brand new chip outperforms slow chip from 10 months ago!
The Z2460 was launched back in January [1], though phones with it have been very slow to appear.
1 - http://www.theverge.com/2012/1/10/2697790/intel-details-new-...
The Z2460 was launched back in January [1], though phones with it have been very slow to appear.
1 - http://www.theverge.com/2012/1/10/2697790/intel-details-new-...
A6 still outperforms the new one:
http://www.engadget.com/2012/09/18/motorolas-razr-i-benchmar...
Also, it's important to note that JS engine optimizations play a huge role in this test, so it's not all just hardware performance, but that doesn't necessarily mean Intel has a disadvantage here. I believe Intel has been working just as heavily to tweak the JS performance in the phones that shipped with their processor.
Also, Intel is being misleading about having a "2 Ghz processor". The thing is it will use that frequency when it's doing the benchmark, so you'll be like "wow, great performance from this chip", but then in 99% of the situations, you'll be using it at its "real frequency" of 1.6 Ghz (Anand says it's still 1.3 Ghz, Engadget may be in error there), so the real performance is probably much worse than the benchmakrs make it look.
http://www.engadget.com/2012/09/18/motorolas-razr-i-benchmar...
Also, it's important to note that JS engine optimizations play a huge role in this test, so it's not all just hardware performance, but that doesn't necessarily mean Intel has a disadvantage here. I believe Intel has been working just as heavily to tweak the JS performance in the phones that shipped with their processor.
Also, Intel is being misleading about having a "2 Ghz processor". The thing is it will use that frequency when it's doing the benchmark, so you'll be like "wow, great performance from this chip", but then in 99% of the situations, you'll be using it at its "real frequency" of 1.6 Ghz (Anand says it's still 1.3 Ghz, Engadget may be in error there), so the real performance is probably much worse than the benchmakrs make it look.
> Also, it's important to note that JS engine optimizations play a huge role in this test
Running the bench on a 4S running iOS6 should fix that.
According to the article (and the original source on anandtech), the actual insight provided by SunSpider is its stressing of the memory interface (this was apparently Intel's original explanation for why Atom crushed A9-based SoC in SunSpider even though the raw processing power was pretty similar), and is a datapoint in support of the hypothesis that Apple's custom core design put a high emphasis on on improving the memory subsystem (and much higher than the stock ARM designs).
Running the bench on a 4S running iOS6 should fix that.
According to the article (and the original source on anandtech), the actual insight provided by SunSpider is its stressing of the memory interface (this was apparently Intel's original explanation for why Atom crushed A9-based SoC in SunSpider even though the raw processing power was pretty similar), and is a datapoint in support of the hypothesis that Apple's custom core design put a high emphasis on on improving the memory subsystem (and much higher than the stock ARM designs).
I agree about the software implication in these benchmarks. It is simply not the same OS, nor browser, nor javascript engine, so I don't see how this test is relevant...
This is also JSC vs V8 (and V8's x86 backend is very mature and well optimized; JSC's ARMv7 backend is pretty good, but nobody's ARM backend has had the same time and effort poured into it as their x86 backends).
The title is misleading because the Intel Atom is Medfield, which is a chip (hurriedly) designed for mobile use from a laptop chip by Intel and hence low power. Anyone know how it would do with a real laptop Atom processor?
While I understand you point, it's rather thin - they are comparing the newest smartphone chip from apple with the newest smartphone chip from intel. Intel brought their best low power engineering to the table - haswell won't be available until 2Q13 which is pretty much an entire phone design cycle. Power doesn't mean much anymore without power/watt.
Also the medfield in the Lava XOLO X900(w/HT) is a single core processor and not a dual core like the A6
Never mind that Sunspider tests the JS engine (which isn't identical on either device) just as much as the processor. Testing JS->engine->OS->processor is a poor method of gauging performance of the processor itself.
That said, I'm surprised no one's attempted a port of Nitro to Android/NDK for benchmarking purposes, to remedy this somewhat.
That said, I'm surprised no one's attempted a port of Nitro to Android/NDK for benchmarking purposes, to remedy this somewhat.
Sunspider is a single-threaded test, so it doesn't matter.
The article fails to mention that there are at least two contributing factors in the technology stack, raw CPU speed and efficiency of the JS engine.
It baffles me that AnandTech, which is so good on so many of these tech topics commits the double flaw of a) using the sunspider benchmark at all (as it's the least meaningful of the various Javascript benchmarks) and b) continually using it to compare devices running different javascript engines (or different versions of the same engine over time) and implying it measures hardware differences.
If you read the original anandtech article (or even just the linked appleinsider repost, which keeps these parts mostly intact), Anand clearly notes that Sunspider is not a very good benchmark these days:
> SunSpider is quickly outlasting its welcome as a smartphone benchmark
but that it keeps significant value in its stressing of the memory interface:
> it does do a great job of highlighting issues with the Cortex A9's memory interface. Intel originally hinted at issues in the A9's memory interface as being why Atom was able to so easily outperform other ARM based SoCs in SunSpider.
He also ends the 3-paragraph note by stating that SPECint would really need to be ported and run to have actionable numbers beyond mere hints.
Your criticism is unwarranted and merely highlights that you haven't bothered reading either the original source or the secondary source linked here.
> SunSpider is quickly outlasting its welcome as a smartphone benchmark
but that it keeps significant value in its stressing of the memory interface:
> it does do a great job of highlighting issues with the Cortex A9's memory interface. Intel originally hinted at issues in the A9's memory interface as being why Atom was able to so easily outperform other ARM based SoCs in SunSpider.
He also ends the 3-paragraph note by stating that SPECint would really need to be ported and run to have actionable numbers beyond mere hints.
Your criticism is unwarranted and merely highlights that you haven't bothered reading either the original source or the secondary source linked here.
Did you read the article? They were perfectly upfront about the limitations of sunspider and only drew the cautious conclusion that the A6 likely has a better cache/memory interface compared to Cortex A9.
I think they're just excited to have a new, unknown chip to try to figure out so any bit on information is worth chewing over.
I think they're just excited to have a new, unknown chip to try to figure out so any bit on information is worth chewing over.
For your second point, the same benchmark on a 4s with ios6 would serve as a control for hardware. Also, considering smartphones are used for web browsing, javascript performance is a valid thing to measure.
According to Weiran, iOS6 on a 4S gets about 2000ms (applying a ~10% improvement to the chart's 2200ms 4S score): https://news.ycombinator.com/item?id=4543358
Replying to various similar comments in one go:
I have read the original AnandTech article, and read perhaps 40 or so very similar articles from them over the last year or so. At no point in that time was the benchmark valid for comparing hardware if you used the stock javascript engine, so it's long, long past "wearing out it's welcome" in that regard.
Any benchmark is suspect, but running a javascript benchmark in order to measure hardware improvements is crazy, as the various javascript developers and phone manufacturers are in a mad scramble to improve on benchmarks and ship the latest code. The resulting graph is almost entirely worthless, you might as well sort them by time of release.
It's only rubbing salt into the wound that they use Sunspider, which even as a javascript benchmark tells you basically nothing about relative javascript performance.
Maybe it does tell you something about memory bandwidth but I can't really take AnandTech's word on that, since they've been using it for something else that it's totally unsuited to measure for so long that they've lost my faith on that topic.
I have read the original AnandTech article, and read perhaps 40 or so very similar articles from them over the last year or so. At no point in that time was the benchmark valid for comparing hardware if you used the stock javascript engine, so it's long, long past "wearing out it's welcome" in that regard.
Any benchmark is suspect, but running a javascript benchmark in order to measure hardware improvements is crazy, as the various javascript developers and phone manufacturers are in a mad scramble to improve on benchmarks and ship the latest code. The resulting graph is almost entirely worthless, you might as well sort them by time of release.
It's only rubbing salt into the wound that they use Sunspider, which even as a javascript benchmark tells you basically nothing about relative javascript performance.
Maybe it does tell you something about memory bandwidth but I can't really take AnandTech's word on that, since they've been using it for something else that it's totally unsuited to measure for so long that they've lost my faith on that topic.
The Anandtech article on Sunspider is mostly just a followup to their article on Geekbench. We'll see an in-depth article once they manage to actually get one of the new iPhones in their lab.
http://www.anandtech.com/show/6298/analyzing-iphone5-geekben...
http://www.anandtech.com/show/6298/analyzing-iphone5-geekben...
What would you suggest as a better Javascript benchmark? V8? Octane?