Google Nexus 4 and Nexus 10 Performance Preview
by Anand Lal Shimpi & Brian Klug on November 2, 2012 11:00 AM ESTEarlier this week Google announced two new flagship Nexus devices: the Nexus 4 smartphone and the Nexus 10 tablet. We received review samples of both earlier this week, and while we're hard at work at full reviews of the devices we couldn't help but share all of the test data we've been able to amass at this point.
For those who aren't familiar with it, the Nexus 4 features Qualcomm's Snapdragon S4 Pro SoC - a quad-core 28nm Krait CPU with Qualcomm's next-generation Adreno 320 GPU. The combination proved quite formidable in the MDP/T we tested, as well as LG's recently announced Optimus G. The SoC drives a 4.7-inch 1280 x 768 IPS display and is paired with 2GB of LPDDR2 memory. The Nexus 4 ships unlocked with 8GB of NAND for $299 without a contract ($349 for the 16GB version). Pair that with DC-HSPA+ support and you get an absolute killer smartphone for use on T-Mobile: no contracts, very low monthly fees, and compelling cellular performance:
Brian will talk more about the combination in his full review, but rest assured that the lack of LTE is workable depending on T-Mobile coverage where you live/travel to.
The Nexus 10 also boasts a brand new SoC: Samsung's Exynos 5 Dual. The Exynos 5 Dual features two ARM Cortex A15 cores running at 1.7GHz as well as ARM's own Mali-T604 GPU. This happens to be the exact same platform used in the new Chromebook, just running Android. The Nexus 10 features a 10.1-inch 2560 x 1600 display, giving it the same resolution as the 13-inch MacBook Pro with Retina Display - but in an even smaller form factor. Google is also aggressive on Nexus 10 pricing: the 16GB WiFi-only tablet sells for $399, with the 32GB version going for $499.
Both Nexus devices run Android 4.2 and are guaranteed to be the first devices to be updated to upcoming Android revisions for the foreseeable future (it's the power of Nexus).
We haven't had a ton of time to test the devices and put this together so you're going to see combined performance charts throughout the rest of this article.
CPU Performance
The big story when it comes to CPU performance is a look at how the Cortex A15s perform under Android. Unfortunately we're still left with mostly browser based benchmarks to measure CPU performance, which actually highlights a major issue in our testing: Android V8 optimization doesn't seem to be anywhere near as good as it is under Chrome OS or Windows. As a result, all of the Nexus 10 performance scores end up slower than the new Chromebook - despite using the same SoC and running Chrome on both platforms. It's also possible that the Exynos 5 Dual in the Chromebook is allowed to burn a bit more power, translating to better performance, but either way the solution here in the Nexus 10 doesn't look as good across the board.
SunSpider performance is good, but not significantly better than Qualcomm's Krait based Snapdragon S4. Both the iPhone 5 and RAZR i are able to outperform the Nexus 10. The S4 Pro based Nexus 4 tends to be in line with other S4 based devices - SunSpider doesn't really give much credit to the extra 2 cores.
BrowserMark puts the Nexus 10 behind many platforms that should be faster, I'm even wondering here if there's some hard partitioning of memory bandwidth between the CPU and GPU to drive the 2560 x 1600 display that's simply choking the CPU here.
The Nexus 4 does ok, but again there seem to be some V8 optimization issues at work here under Android 4.2. At 1.5GHz it should deliver at least the performance of the dual-core Snapdragon S4 solutions.
Octane is the first test where the Cortex A15s are really able to flex their muscle - the Exynos 5 Dual based Nexus 10 manages to outperform the RAZR i by 34%, and compared to the A6/Swift based iPhone 5 the advantage grows to 64%.
The Nexus 4 performs about in line with other Snapdragon S4 based devices, although once again the extra 2 cores don't seem to be doing much for it here at all.
Kraken also paints the Cortex A15 based Nexus 10 in a good light: there's a 30% advantage over the RAZR i and a 76% advantage over the iPhone 5. These numbers will shrink a bit compared to other tablets, but not by much. The Nexus 4, once again, ends up performing similarly to dual-core Snapdragon S4 based devices.
Overall, the Nexus 10 results show us some real promise for what we can expect from ARM Cortex A15 based SoCs. The potential upside to this new architecture is huge.
244 Comments
View All Comments
at_sucks - Friday, November 2, 2012 - link
I don't understand why in review mobile phones and tablet performance has to be compared.. Also in Sunspider and browser test IPad performance is cleverly omitted .. And also if mobile phones has to be included, include best selling andriod phones for each benchmark tests..Check this out, not very detailed nexus 10 review but do justice
http://www.zdnet.com/google-nexus-10-review-700000...
Really reading this review lost total faith on AT. Looks like Anand lost his credibility.
UpSpin - Friday, November 2, 2012 - link
Same thoughts. The comparison is misleading and below AT standard.Sure, the Nexus 4 results don't look good, even if you make a fair chart, but they look that bad that something must be wrong. Because AT only used Browser benchmarks, it must be a software issue!
If the LG Optimus G has the same hardware, yet scores much higher in 3D benchmarks, then there's something wrong. If the Optimus G scores higher because the bencmark crashes, then it's unfair to include the benchmarks at all if you run it differently on other phones.
With this preview AT lost a lot of credibility because of their inconsistent comparisons and benchmark runs. I hope you'll fix it and also contact Google why the browser!! (not the CPU as you want to tell us) performs so bad on the Nexus 4.
ratte - Friday, November 2, 2012 - link
If you actually READ the article it's explained.hint: "thermal throttling "
andrewaggb - Friday, November 2, 2012 - link
I'd like a more well rounded benchmark suite.If there really isn't any cross-platform benchmarks, consider getting some written.
I don't take browser benchmarks seriously, at least not javascript ones. Stuff like page load time is real, but everybody and their dog is optimizing javascript right now. It's impossible for me to tell the speed of the hardware when the version of the javascript engine is probably different on each platform tested. That's not a good benchmark for comparing cpu's, though it may be fair for comparing phones/platforms on the whole.
I think stuff like app start time would be nice. I think all platforms have angry birds space (even surface), all of them have mail readers, browsers, etc.
Task switching performance, wifi file transfer speed. Seems like there's more that can be tested.
Number of steps and average time for an experienced user to perform basic operations (unlock, connect to wifi, install an app, write a text message, check weather, facebook etc.) etc. That kind of stuff would be nice. Also I see battery life tests, but never charge time tests.
DLNA/smartglass/apple tv type stuff, how cleanly does it work to send a media file, web link, youtube, etc to a tv or other device, and to get it back again or play to tablet type options.
Even a giant checklist of stuff you'd probably want to do and run, with how long it takes you to do it, and how quickly the device can do it, would be nice.
andrewaggb - Friday, November 2, 2012 - link
"If there really isn't any cross-platform benchmarks, consider getting some written."I mean cross platform cpu benchmarks
yyrkoon - Friday, November 2, 2012 - link
You need to understand.Browsers, and hardware designers are all starting to ( have been for a while ) optimize for javascript performance.
The reason is simple. Any browser application that is doing anything serious will very likely be using javascript, and lot of it. Games, or regular applications.
Technically, the javascript engine is in the browser used. However, as stated above. Everyone should be optimizing their hardware for javascript performance. If they're not, or being lazy about it. Then they are just wrong.
Web apps are about as cross platform as it gets with mobile devices. With any semblance of ease, that is.
yyrkoon - Friday, November 2, 2012 - link
Oh, and in case it is not already obvious.By extension, any reviewer who does not benchmark javascript, is also wrong.
andrewaggb - Friday, November 2, 2012 - link
GL Benchmark is written in c/c++ with platform dependant code for the ui. That's a common theme for many apps.Apps with serious performance requirements are going to be written in c/c++. Especially in mobile where you have slow cpu's. Javascript is important, but it's not the end all be all.
Javascript is just like java, .net, python, php, and a whole host of other 'higher level' programming languages. It tends to not be as fast as c/c++. If you're application is cpu bound, you should be using c/++.
If you're really interested,
http://shootout.alioth.debian.org/u64/benchmark.ph...
MadMan007 - Saturday, November 3, 2012 - link
This x100. Give me real-world use scenarios, or at LEAST translate how 2x in a benchmark translates into real-world use. Does it seem that much faster, is the bottleneck elsewhere so it doesn't matter anyway, or whatever.UpSpin - Friday, November 2, 2012 - link
Why do you call browser benchmarks as CPU benchmarks? You also don't compare the performance of different Intel CPUs by comparing the Sunspider results of IE with them done in Safari on a Mac.Browser benchmarks are interesting, but please call them what they are: Browser benchmarks, not more not less. If you use a different browser, you'll get different results, with the same SoC. Thus name the Android version, browser version and Smartphone you used, everything else is misleading.
If you want to compare CPU speed, which I highly recommend, then do some single and multi-core number crunching with specific apps, just as it gets done on regular computer reviews, too.