Apple A6 SoC outperforms Intel Atom in JavaScript test
A test of the iPhone 5's new A6 processor yielded a surprisingly quick score on SunSpider's JavaScript benchmarking tool, with the system on chip topping even the Medfield Atom Z2460.
While not an official test, AnandTech was able to run SunSpider v0.9.1 on an iPhone 5 review sample on Wednesday, and watched as the dual-core A6 chip chewed through the benchmark in 914.7 milliseconds, outpacing the second place Lava XOLO X900's time of 1279.4 milliseconds. The Lava XOLO runs a single-core Intel Atom Z2460 Penwell SoC built on the chip maker's Medfield platform.
Source: AnandTech
As noted by the publication, SunSpider is slowly becoming outdated as a smartphone benchmark, however it highlights issues pertaining to ARM's Cortex A9 memory interface.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," reports Anand Shimpi.
In a Monday report, online benchmarking site Geekbench reportedly logged an iPhone 5's performance data, outing the specifications of the handset's A6 chip. At the time, it was thought that Apple made dramatic improvements to the standard ARM core's memory interface as the Geekbench data suggested a drastic improvement from previous Cortex A9 chips.
Although it hasn't yet been confirmed, the A6 chip is reportedly Apple's first attempt at designing an ARM core in-house.
While not an official test, AnandTech was able to run SunSpider v0.9.1 on an iPhone 5 review sample on Wednesday, and watched as the dual-core A6 chip chewed through the benchmark in 914.7 milliseconds, outpacing the second place Lava XOLO X900's time of 1279.4 milliseconds. The Lava XOLO runs a single-core Intel Atom Z2460 Penwell SoC built on the chip maker's Medfield platform.
Source: AnandTech
As noted by the publication, SunSpider is slowly becoming outdated as a smartphone benchmark, however it highlights issues pertaining to ARM's Cortex A9 memory interface.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," reports Anand Shimpi.
In a Monday report, online benchmarking site Geekbench reportedly logged an iPhone 5's performance data, outing the specifications of the handset's A6 chip. At the time, it was thought that Apple made dramatic improvements to the standard ARM core's memory interface as the Geekbench data suggested a drastic improvement from previous Cortex A9 chips.
Although it hasn't yet been confirmed, the A6 chip is reportedly Apple's first attempt at designing an ARM core in-house.
Comments
"At the time..." that was then... so what do they think now?
Quote:
Originally Posted by Douglas Bailey
"At the time, it was thought that Apple made dramatic improvements to the standard ARM core's memory interface as the Geekbench data suggested a drastic improvement from previous Cortex A9 chips."
"At the time..." that was then... so what do they think now?
They think it's apple's own in-house arm design now, and not based off of the cortex a9 or a15, but it's not confirmed.
They should have added both 4S iOS 5 and 4S iOS 6 results as any speed increase in sunspider cannot be attributed to the SoC alone.
Quote:
Originally Posted by AppleInsider
outpacing the second place Lava XOLO X900's time of 1279.4 milliseconds. The Lava XOLO runs a single-core Intel Atom Z2460 Penwell SoC built on the chip maker's Medfield platform.
So as per the previous discussion with Geekbench, the single core Atom actually does quite well compared to the dual core A6.
My own opinion is that Ghz and core count is irrelevant, performance / Watt is the only relevant number for these kinds of devices and the iPhone5 is a winner in that regard.
Quote:
Originally Posted by mausz
So as per the previous discussion with Geekbench, the single core Atom actually does quite well compared to the dual core A6.
In difference to Geekbench Sunspider does only utilize 1 Core.
Quote:
Originally Posted by smalM
In difference to Geekbench Sunspider does only utilize 1 Core.
How can you force the use of 1 cpu core from javascript ? You cannot as javascript has no notion of cpu cores or affinity.
I know most javascript engines use only 1 core (although IE uses multithreading for compiling and running) but in speculation maybe the biggest change with ios6 is the new javascript engine in mobile safari which uses both cores.
That's why the performance with a 4S with ios 6 is relevant. Googling found a few results which show the 4S went from 2200 to 1800 with ios6, so 18% improvement is due to ios6, So iPhone5 with ios5 (not correct to extrapolate this but still) would probably have scored 1165 in sunspider. Still good.
Quote:
Originally Posted by smalM
In difference to Geekbench Sunspider does only utilize 1 Core.
Can you then explain why 4-core international S3 is significantly faster than US (AT&T) version with 2-cores ?
Quote:
Originally Posted by smalM
In difference to Geekbench Sunspider does only utilize 1 Core.
So how does that explain the variation in the Galaxy S III scores, between quad core with 1GB of RAM, dual core with 2GB RAM, etc?
Quote:
Originally Posted by hill60
So how does that explain the variation in the Galaxy S III scores, between quad core with 1GB of RAM, dual core with 2GB RAM, etc?
As a general point, people often state multi cores are not handy because each process is single-threaded and most mobile apps do not employ multithreading. However, with multiple cores processes can be run on different cores. So a single core javascript benchmark can run on one, and for instance whatsapp and other multitasking processes can run on other cores. So with multiple cores and single-threaded apps you will see better performance because the processes do not (or less) hinder eachother.
Wether this explains the large differences.... I know android took a big leap in sunspider scores when the chrome browser arrived to take the place of the stock android browser. Also large differences have been seen in 3.x,4.0.x en 4.1.x. As Anandtech does not state this information it's hard to tell what's going on here.
Quote:
Originally Posted by Constable Odo
Jeez. That Nokia Lumia 900 is a snail. I sure hope the Lumia 920 does a lot better.
Absolutely, the new Lumia 920 has the Qualcomm S4, and the new wp8 ie10 browser, so should do a lot better (I hope for them
LOL
The truth is I suspect that it was more a case of the Jobsian 'keep it simple stupid' philosophy in action. Apple long ago decided talking too technically is a bad way to go.
LOL, I bet a few cheepo PC owners must be thinking that.
Quote:
Originally Posted by jason98
Can you then explain why 4-core international S3 is significantly faster than US (AT&T) version with 2-cores ?
As was explained above, SunSpider is not the only process running. The more cores you have OTHER processes can be spread around freeing up CPU time for SunSpider to run. We also don't know what other processes are running on AT&T's phones. Just look at T-Mobile. Why is their S3 faster?
If the test was multi-core aware, then theoretically having twice as many cores would cut the time in half, or at least approach it.
Another factor could be the clock speed. Maybe the international S3 is clocked higher than the US model?
Quote:
Originally Posted by mjtomlin
As was explained above, SunSpider is not the only process running. The more cores you have OTHER processes can be spread around freeing up CPU time for SunSpider to run. We also don't know what other processes are running on AT&T's phones. Just look at T-Mobile. Why is their S3 faster?
If the test was multi-core aware, then theoretically having twice as many cores would cut the time in half, or at least approach it.
Another factor could be the clock speed. Maybe the international S3 is clocked higher than the US model?
I doubt that Anand did not take care of eliminating influences of "other" apps while running these tests. As for the frequencies, I believe they both have 1.4GHz.
Seems like browser tests are utilizing multi-cores somehow but they are not type of scalable linearly.
Quote:
Originally Posted by mjtomlin
As was explained above, SunSpider is not the only process running. The more cores you have OTHER processes can be spread around freeing up CPU time for SunSpider to run. We also don't know what other processes are running on AT&T's phones. Just look at T-Mobile. Why is their S3 faster?
If the test was multi-core aware, then theoretically having twice as many cores would cut the time in half, or at least approach it.
Another factor could be the clock speed. Maybe the international S3 is clocked higher than the US model?
And just to add another factor. I've read reports which have suggested newer ARM cores could employ turbo modes just like Intel desktop processors. So when a benchmark runs and only stresses one core, this core gets 'overclocked' as long as its within the thermal evelope.
This is one of the issues I have with some benchmarks. Sometimes a SoC uses more power while stressed but it's doing the work faster as well so in practice it would use more power but for a shorter amount of time. The net effect should be looked at.