iPhone 5/A6 SunSpider Performance: Faster than Intel's Atom Z2460
by Anand Lal Shimpi & Brian Klug on September 19, 2012 12:49 AM EST- Posted in
- Smartphones
- Apple
- Mobile
- iPhone 5
The first iPhone 5 reviews have lifted, confirming the leaked Geekbench data we saw in our earlier post. Apple's A6 appears to feature two custom ARM cores running at up to 1GHz. A new datapoint comes courtesy of our own Brian Klug who's currently visiting LG in Seoul, South Korea. He ran into Vincent Nguyen of Slashgear fame, who kindly let him run SunSpider 0.9.1 on Vincent's iPhone 5 review sample. The score? 914.7ms.
SunSpider is quickly outlasting its welcome as a smartphone benchmark, but 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. As we surmised in our A6 Geekbench post, it looks like Apple specifically targeted improvements in the memory subsystem when designing the A6's CPU cores. The result is the fastest SunSpider test we've ever recorded on a smartphone - faster even than Intel's Atom Z2460.
This doesn't tell us much about the A6's architecture other than it's likely got a better cache/memory interface than ARM's Cortex A9. What we really need is for someone to port SPECint to iOS...
110 Comments
View All Comments
Penti - Wednesday, September 19, 2012 - link
This is on slightly older stuff, what about Z2460 on 2GHz and newer Webkit? Should be close, and this Atom is still the same architecture as in 2008. It's a lot faster then the A5 or even A5X. If the clock speed is correct it should be the fastest ARM-around, but software matters so much here. Even at twice the GPU power and very fast custom CPU it's not a chip that will be miles ahead of everything for long. Much of the performance have probably been from faster memory and it's not a next generation GPU, even if it should fit Apple very well it's not like they will smoke next generation chips in everything or always have the best software.Current comparisons against other fast devices with updated software will be interesting, but I will wait for the full review for that. Compared to a 1.6GHz single-core Atom on a not so new old build of 2.3 it should damn well be faster though. Though it's fun it's basically Sony PS Vita power in your phone. To bad iPod Touch with A5 has no chance in competing against portable game consoles though, due to price. Would be interesting to see more hard-core stuff in that department.
Wilco1 - Wednesday, September 19, 2012 - link
The Sunspider score for Z2460 at 2GHz turbo is 1062 (http://www.engadget.com/2012/09/18/motorolas-razr-... So the A6 beats the only benchmark where Atom looked good by a big margin. And that for the very latest Atom... This is the final nail in the coffin for Atom - it'll be late 2013/early 2014 before the next generation will appear.The A6 will be king of single-threaded performance until Cortex-A15 comes out in the next few months - at 2GHz it should double single-threaded performance again!
j85 - Wednesday, September 19, 2012 - link
Most of the improvements in SunSpider have come from having the latest browser updates. For example, my Galaxy Nexus scores 1413ms on SunSpider in Chrome, which is 400ms less than stock. iOS6 naturally has the latest version of WebKit. The Droid Razr with the Atom Processor is not running the latest version of Android nor was it running Chrome.Penti - Wednesday, September 19, 2012 - link
It still looks good in Sunspider compared to other Android phones, I've never read anything else into it than that. I wouldn't buy it over fast Quadcore ARM Android-powered units because of it's speed, the devices are in totally different categories. Nice to see what they can do with a single-core Atom though. Glad to see a benchmark though, might have glanced over others without taking notice too.It's not the latest Atom however, it's still the 2008-era Bonwell architecture. It is the latest product, not architecture. Nobody expects early SoC's/platforms of doing well and it seems to do fine with a ~5Wh battery. From how it looked back in 2008 it's pretty impressive. That's the same year Cortex-A8 SoC's came available. Next generation is in the pipeline and nobody expected the Z2460 to be in every high-end device up until then. But it's not like you can run Android on the A6 either. It competes against other low-end SoC's. You can always use a larger variant on say a tablet or other type of devices. Or you can stop caring about what architecture it is or you have to run and simply build good products. Atom/PowerVR might not be my favorite combination, but it has it's uses.
Wilco1 - Wednesday, September 19, 2012 - link
Atom does well on Sunspider precisely because of software optimizations. It's the only benchmark it used to win. ARM based phones are catching up with similar optimizations - there is no doubt that iOS6 has had significant JS tuning done, and future versions of Android will add more as well (a score of 1400 with stock browser is now becoming typical).Yes the Z2460 is based on the original Atom architecture with only minor changes. But it is actually the very first Atom chip that is capable of being a smartphone CPU, as the original Atoms weren't SoCs and used way too much power. If Intel had released a Z2460 equivalent in 2008, they could have gained a lot of the smartphone market. Now they have to battle with much faster and powerful ARM cores...
Penti - Thursday, September 20, 2012 - link
Of course it's a low end chip that only does well in this benchmark. Only having a PowerVR 540MP2-gpu is abysmal now and it doesn't have that much memory bandwidth. At least TI have released OMAP4470 to get better gpu-performance. Until they have A15's out. It's still impressive to see a chip pushed or that you can update/optimize even better then what they did on 2.3.7/older Webkit.Atom still has the same architecture as in 2008, roughly the same clocks, ARM-cores doesn't it isn't just power reductions, multi-core, memory bandwidth, die-shrinks (65nm to 28/32nm) it's also new architecture. It's not like it looks worse and worse for them. Intel already have had success in this field with the XScale line that they sold off. It's not like they need to take the entire market, it's impressive to be there at all regarding how fast ARM-chips have developed just in the last few years. They are large superscalar branch prediciting out-of-order chips already. They have a much better place in the Android ecosystem then MIPS at least. You need to be pretty flexible and move pretty fast to stay there, at least if you don't have a robust architecture to compete with. In 2008 most stuff was still on ARM11.
vdx660 - Wednesday, September 19, 2012 - link
The funny thing here is that the authors left out a glaring detail-- that Apple A6 has a Dual Core vs an Intel Medfield Single Core.darkcrayon - Wednesday, September 19, 2012 - link
They should've included it but I wouldn't call it "glaring". If the single core is what's available and has a similar power profile to the A6, then why shouldn't they be compared? It's a completely different architecture. It's not like the Medfield phone they tested was a dual core, but they intentionally disabled one of the cores to perform the test. Just like the A6 should be compared directly to the quad core Exynos in one of the Galaxy S III's.tipoo - Wednesday, September 19, 2012 - link
Exactly what darkcrayon said, core count and clock speed are irrelevant next to power draw and performance. The single core Atom there drew as much or more power, so it is comparable.Wilco1 - Wednesday, September 19, 2012 - link
Sunspider is a single threaded benchmark so the number of cores (and hyperthreading) does not matter here.