Apple's iPad A4 SoC: Faster than the Nexus One's Snapdragon?by Anand Lal Shimpi on April 3, 2010 3:52 PM EST
- Posted in
Fresh out the frying pan and into the fire, I just finished my Nexus One review late last night only to have my iPad preorder show up early this afternoon. I had been preparing for it's arrival not by downloading apps but by figuring out what comparative benchmarks I wanted to run on the iPhone 3GS and Nexus One.
As the first device to use Apple's A4 SoC I wanted to see how it stacked up against the Cortex A8 and Qualcomm's QSD8250. All three chips appear to be dual issue in order architectures with varying pipeline depths, clock speeds and cache sizes.
At 600MHz the Cortex A8 in the iPhone 3GS is the slowest out of the bunch. The Snapdragon is much faster as we just established thanks in part to it's 1GHz clock speed. But what about Apple's 1GHz A4?
There's very little we know about the A4 other than it's operating frequency. It is manufactured by Samsung but on an unknown process node. Jon Stokes recently stated that Apple's secrecy surrounding the chip is because it isn't anything special, just a Cortex A8. If that is true, I suspect that it would have to be manufactured at 45nm in order to reach such a high clock speed.
With a new silicon mask there's also the chance that Apple moved to LPDDR2 to boost memory bandwidth; a change that most SoC makers are planning to make this year.
So how does Apple's A4 stack up against today's favorite smartphone brainchild? Keep in mind that these results are generated by running two different OSes (Android 2.1 and iPhone OS 3.2) and two different browsers. What we're looking at is the performance delivered by the combination of the CPU and the software stack:
|Applications Processor Performance|
|Apple iPad (Apple A4)||Apple iPhone 3GS (ARM Cortex A8)||Google Nexus One (Qualcomm Snapdragon QSD8250)||% A4 Faster than Snapdragon|
|Load www.anandtech.com||6.2 seconds||9.3 seconds||8.8 seconds||41%|
|Load www.digg.com||10.6 seconds||18.0 seconds||11.5 seconds||8.7%|
|Load www.tomshardware.com||7.9 seconds||13.9 seconds||8.6 seconds||8.7%|
|Load www.arstechnica.com||7.8 seconds||13.8 seconds||11.0 seconds||39.9%|
|Load www.legitreviews.com||6.8 seconds||12.3 seconds||8.6 seconds||26%|
|Load www.techreport.com||3.7 seconds||7.4 seconds||4.2 seconds||11.6%|
|Load www.engadget.com||13.8 seconds||22.8 seconds||22.0 seconds||59.4%|
|Load www.gizmodo.com||14.1 seconds||21.4 seconds||16.7 seconds||18.5%|
|Load m.cnn.com||3.0 seconds||6.0 seconds||2.6 seconds||-11.8%|
Unless otherwise specified, I loaded the full version of all of the websites above (the exception being CNN, where I used the mobile site). To ensure reliability, I ran all of these tests at least 5 times, threw out any outliers and averaged the rest. The rests were also run at around the same time to ensure that content on the sites was as similar as possible (and thus shouldn't be compared to this morning's Nexus One results). You'll note that the Engadget results are a bit odd. It looks like the iPhone and Nexus One scores are bottlenecked somewhere else (there seemed to be some network issue plaguing the loads, but it wasn't present on the iPad), but if you toss out the very large differences you end up with what I believe to be the real story here. Update: Flash wasn't enabled on any device (not supported on iPad/iPhone, not officially available on Android yet), and all three devices connected to the same WiFi network. The Apple devices used mobile Safari, while the Android device used the Android Browser. Both are WebKit based but there are obvious, unavoidable software differences.
Removing the AnandTech, Ars Technica and Engadget loads (which were repeatable, but unusually long) the iPad loads web pages 10% faster than the Nexus One. If you include those three results the advantage grows to 22.5%. I'd say somewhere in the 10% range is probably realistic for how much faster the A4 is compared to the Snapdragon.
If we take the network out of the equation, the A4 in the iPad has a 37.6% performance advantage over the Qualcomm QSD8250. This actually supports some of the larger performance differences we saw earlier. If Apple can manage to deliver this sort of performance in its smartphone version of the A4, we're in for a treat.
The why is much more difficult to ascertain. It could be as simple as the the iPad OS being better optimized than Android, a definite possibility given how much longer Apple has been working on it compared to Google. The advantage could also be hardware. The A4 may boast higher IPC than Qualcomm's Snapdragon thanks to better core architecture, larger caches or a faster memory bus. The likely case is somewhere in between, where the iPad's advantage comes from a combination of hardware and software.
It could also be a power optimization thing. The A4 in the iPad is paired with a much larger battery than the QSD8250 in the Nexus One, Apple may be able to run the SoC at more aggressive performance settings since it doesn't have to worry about battery life as much. Either way the one thing we can be sure of is Apple's A4 SoC is much more like a 1GHz Cortex A8 rather than anything more exotic. Good work Jon :)
I should note that while the performance improvement is significant, it's not earth shattering. Despite the early reports of the iPad being blazingly fast, I found it just "acceptable" in my limited time with it thus far. I'll go into greater detail in my full review later.
This does bode well for the upcoming 4th generation iPhone, which is widely expected to also use the Apple A4 SoC. That upgrade alone should put the next iPhone ahead of Google's Nexus One in performance, assuming that it offers the same performance as it does in the iPad. Pair it with a modernized and feature heavy iPhone OS 4.0 and we might see an Apple answer to Android in 2010.
The A4 is particularly exciting because it combines Snapdragon-like CPU performance with a PowerVR SGX GPU. A much better option than the aging ATI core used in Qualcomm's QSD8x50 series.
With Apple showing its A4 performance this early, Qualcomm also has a target to aim at. The first single-core 45nm Snapdragon SoC due out in 2010 will run at 1.3GHz. That could be enough to either equal or outperform Apple's A4 based on what we've seen here today.
Expect our full review of Apple's iPad as well as more discussion about the A4 next week. Have a great weekend guys.
Post Your CommentPlease log in or sign up to comment.
View All Comments
metafor - Sunday, April 4, 2010 - link"Also, Snapdragon may claim their core to be capable of 1 GHz, but most of the time it is going to be underclocked to save on battery life. Can you repeat the tests with both the Snapdragon products connected to a power outlet? (Hopefully Qualcomm was prudent enough to let the chipset run at full speed when in operation while being charged)."
I don't believe cell phones work like that. They don't have different power profile modes like laptops do as they're almost never expected to be used while charging. And for the most part, the actual applications processor has no knowledge of "battery vs hooked up" on a hardware throttling level.
Android may throttle the processor more aggressively than the iPhoneOS. But we'll never know.
leatherseat - Saturday, April 3, 2010 - linkYou are comparing apples and oranges. The browsers are different. The support for flash is different. I would hazzard a guess that m.cnn.com is a more accurate overall speed comparison, because of the lack of flash on that particular website.
You should also show the connection, since I'd take it the ipad is based on wifi, while the phones could be both.
It pains me to see when anandtech is reviewing intel/AMD processors they go to the highest extent to keep most of the factors the same, while revewing anand's beloved apple products, it jumps into conclusion so fast, without proper experimental design.
Your results do not show that A4 is better or worse than the snap dragon. It only shows that ipad browsing is faster than the nexus one browsing, under your testing conditions.
Anand Lal Shimpi - Saturday, April 3, 2010 - linkMy apologies for not making this clear earlier - flash wasn't enabled on any platform and all three used WiFi, connected to the same access point from the same physical distance (all 4.5 feet away with line of sight, no obstructions).
Sapient2007 - Sunday, April 4, 2010 - linkhi Anand,
the difference between snapdrAgon and cortex A8 is not just clock speed. To me it SeeMs that the architectural differences Qualcomm employed to get snapdragon to get to 1GHz may be partly responsible. do you have access to an over clocked droid running at 1 GHz or an OMAP 3630 based device? So that we may check other 1GHz cortex A8 processors.
Crowl - Sunday, April 4, 2010 - linkWhat speed was the wifi connection used, were you limiting the ipad to the same wireless g supported by the other two or was it connecting as a wireless n device?
ltcommanderdata - Saturday, April 3, 2010 - linkhttp://www.ifixit.com/Teardown/iPad-Teardown/2183/...
iFixit's teardown show the iPad is using 2Gb RAM chips just like the third-gen iPhone and iPod Touch. However, they conclude that the iPad is using 2 of these chips for 512MB total. This should be ideal for the multitasking support hopefully coming in iPhone OS 4.
The K4X2G643GE RAM part number also seems to indicate that the iPad has a 64-bit memory bus compared to the 32-bit memory bus I believe you said was used in the iPhone and other smartphones so that could definitely be contributing to the speed increase.
Anand Lal Shimpi - Saturday, April 3, 2010 - linkThose two chips are actually MLC NAND devices, not RAM. The DRAM is actually located on the A4 package itself. ifixit has apparently x-rayed the chip and revealed that it's got 512MB of on-package DRAM, identical to Qualcomm's QSD8250.
ltcommanderdata - Saturday, April 3, 2010 - linkThe K4X2G643GE part number that iFixit and I are referring to is printed on the A4 chip itself presumably because it is integrated into the package as you say and refers to the DRAM. The two separate NAND flash chips have a K9LCG08U1M part number.
Anand Lal Shimpi - Sunday, April 4, 2010 - linkMy mistake! I thought you were referring to the NAND devices that ifixit originally referred to as DRAM. Two devices doesn't necessarily mean it's a 64-bit wide interface. I believe such a wide memory interface is still unheard of in a smartphone class SoC.
ltcommanderdata - Sunday, April 4, 2010 - linkIn the end it looks like I was mistaken afterall in quoting iFixit on 512MB of RAM since they've since changed their interpretation of 2Gb to refer to both RAM chips together for 256MB of RAM total. This reportedly matches software developer feedback finding available free memory on the iPad being similar to the iPhone 3GS.
I did post over at iFixit to see what they make of the 64-bit memory bus question since I believe in Samsung's nomenclature the 6th and 7th character in the K4X2G643GE part name refers to interface width.
Samsung's Mobile RAM brochure only mentions up to 32-bit organization but I guess the 2 RAM chips could be arranged in a dual channel configuration for 64-bit width. Doubled memory bandwidth would be some consolation for the lack of doubled memory amount and would probably be most helpful to the GPU.
And thank you for the time you spend in responding to readers comments. It's definitely appreciated and adds to the quality of the articles and website.