Think multi-core chipsets are a gimmick? Real-life Android study proves all cores matter

posted by Daniel P. / Sep 03, 2015, 6:43 AM

Think multi-core chipsets are a gimmick? Real-life Android study proves all cores matter

Why have octa-core and even deca-core mobile chipsets if there are barely any apps out there that can take advantage of multithreading with more than two to four processes at once - that's been the conventional logic so far, right? Well, the geeky bunch from Anandtech set on a massive crusade to settle once and for all the multi-core performance debate for everyday scenarios you do with your Android phone.

They picked a few common tasks, like opening Hangouts and writing a message, rendering a website, running the camera, installing apps from the Play Store, playing games, and so on. The goal was to determine if and where multi-core matters, and if it matters more in big.LITTLE configs, like the one on many octa-core chipsets at the moments, that combine powerful but battery-hungry Cortex-A57 cores with humbler, but frugal A-53s for the more mundane tasks. They also threw in other designs, like MediaTek's Helio X10, which combines eight lowly A53s in the so-called symmetric design.

Their analysis is a thorough 16-pager, full of jargon and graphs for each usage scenario, but long story short - the more cores, the merrier, if load distribution and power management are done right. A cluster of four cores was more efficient than one with two, two separate clusters with small and big cores fared better than one heterogenous cluster, and so on. While most usage scenarios showed that four cores is the optimal variant at the moment for Android's multithreading universe, there were certain scenarios where all eight cores made a difference, such as website rendering and app installation. Given how often we use the browsers on our handsets, it's reassuring to know that octa-core setups aren't just a marketing gimmick, but there's actual use for them in real-life scenarios. The key conclusion?

In the end what we should take away from this analysis is that Android devices can make much better use of multi-threading than initially expected. There's very solid evidence that not only are 4.4 big.LITTLE designs validated, but we also find practical benefits of using 8-core "little" designs over similar single-cluster 4-core SoCs.

Multicore mobile chipsets make sense for Android

source: Anandtech

FEATURED VIDEO

Options

24 Comments

1. rd_nest

Posts: 1656; Member since: Jun 06, 2010

Leave it to AndreiLux to explain on ARM/Samsung designs. He's among the best devs for Exynos devices in XDA. His in-depth article on Exynos 7420 is a must read.

posted on Sep 03, 2015, 7:05 AM 13

4. vincelongman

Posts: 5745; Member since: Feb 10, 2013

Do you have a link for his Exynos 7420 article? I can only find his Exynos 5410 article (great article BTW)

posted on Sep 03, 2015, 7:43 AM 0

7. rd_nest

Posts: 1656; Member since: Jun 06, 2010

^^http://www.anandtech.com/show/9330/exynos-7420-deep-dive

posted on Sep 03, 2015, 8:53 AM 2

13. vincelongman

Posts: 5745; Member since: Feb 10, 2013

Oh I've read that one, brilliant article I didn't realise AndreiLux from XDA was the same Andrei from AnandTech lol

posted on Sep 03, 2015, 4:20 PM 0

23. alex3run

Posts: 715; Member since: May 18, 2014

He knows hardware well but his kernels arent the best.

posted on Sep 04, 2015, 5:13 PM 0

2. Shubham412302

Posts: 586; Member since: Nov 09, 2011

as far as i know when using titanium backup my old phone used to get slow for other operations but for octacore while backing up and restoring you dont even know that titanium backup is running same goes for many app installation from playstore

posted on Sep 03, 2015, 7:07 AM 9

3. jroc74

Posts: 6023; Member since: Dec 30, 2010

As it should be. Almost exactly the way it is on desktops, laptops.

posted on Sep 03, 2015, 7:38 AM 2

5. TylerGrunter

Posts: 1544; Member since: Feb 16, 2012

Well, I must admit that I read the article and I don't 100% agree with the way it was done. The only thing it proves is that the applications in Android are better threaded than most of us expected, most of them being optimized for 4 cores. I will not claim the results are wrong, but it does not answer the real question: Are really useful the extra cores? In other words: do they help to speed up with tasks or to save battery? Sadly the way he tested does not prove either really. Even with that one of the best articles I read in Anandtech in quite a while.

posted on Sep 03, 2015, 7:43 AM 2

8. RoboticEngi

Posts: 1251; Member since: Dec 03, 2014

If they are better threaded, they will be faster.

posted on Sep 03, 2015, 9:30 AM 2

9. TylerGrunter

Posts: 1544; Member since: Feb 16, 2012

Can you PROVE that? Because that's my point.It is all an assumption. And just in case you want to know where I want to go here you have an old article proving that having more threads doesn't necessarily means being faster or using less battery:http://research.microsoft.com:8082/en-us/um/people/zhao/pubs/hotpower13gpu.pdf I know it is an old article, but... Have things changed enough to justify the high number of cores? That's the real question. And the Anandtech article doesn't clear the answer, only that the situation is better than many though, but not HOW MUCH better.

posted on Sep 03, 2015, 10:08 AM 0

12. Niva.

Posts: 440; Member since: Jan 05, 2015

You're missing the point. The cores may help in certain scenarios, in general though he said currently the optimal setup for Android is 4 cores symmetric cluster. The article on anandtech said that, the post above said that. So right now for the general user anything more than 4 cores really doesn't matter much... but there is a "but."

posted on Sep 03, 2015, 12:16 PM 3

15. TylerGrunter

Posts: 1544; Member since: Feb 16, 2012

I'm confused Niva. What point am I missing? I was the one saying that for most task 4 cores is the optimal. Were you trying to reply to RoboticEngi?

posted on Sep 04, 2015, 2:57 AM 0

14. vincelongman

Posts: 5745; Member since: Feb 10, 2013

That Exynos 5410 is very very outdated now compared to the Exynos 7420 (e.g. no HMP and older cores) Also Samsung/Google have improved their software I'd like to see that study again but with next years' SoCs, iOS 10 (iOS 9 brings more use of Metal) and Android N (will probably have Vulkan) *A10 (2x new custom cores?) *820 (4x Kryo cores) *new Intel Atom X7 (4x custom cores) *next Exynos (4x + 4x Mongoose cores?) *Kirin 950 (4x A72 + 4x A53e) *MediaTek Helio X30 (4x A72 + 2x A53 + 2x A53 + 2x A53) And for tablet SoCs *next Tegra (2x new custom cores?) *A10X (3x new custom cores?) *Intel Core m7 (2x Kaby Lake cores with Hyperthreading)

posted on Sep 03, 2015, 4:57 PM 0

16. TylerGrunter

Posts: 1544; Member since: Feb 16, 2012

Just one correction, this year Apple will reveal the A9/A9X, not the A10 yet! Don't be so hasty! And everyone is pressuring Andrei already (look at the comments in the article) so that he does more tests like that. We'll see what he can do, at the end of the day he is also human ;)

posted on Sep 04, 2015, 3:01 AM 0

18. vincelongman

Posts: 5745; Member since: Feb 10, 2013

I actually meant the A10, the SoC in next year's iPhone 7 (hence the meant of iOS 10 as well) Yea true, he's done so many great articles already My list is far too long to be realistic :D

posted on Sep 04, 2015, 3:17 AM 0

19. TylerGrunter

Posts: 1544; Member since: Feb 16, 2012

Man, that SoC is more than year in the future! Can you be happy if he tests the A9/A9X? And I would like him to do a complete different study: - Take Mediatek MT6795, octa-core with HMP - Take 3 or 4 different workloads (Chrome heavy web page, starting one application, play a game, etc) - Test each load with 8 cores, 6 cores, 4 cores, 3 cores and 2 cores (by disabling cores, can be done with root). - Monitor not only the threads but also: how fast the task finished and how much power was used. That's the test that would be relevant for the discussion: Are octa cores actually useful?

posted on Sep 04, 2015, 4:05 AM 0

20. vincelongman

Posts: 5745; Member since: Feb 10, 2013

But that's just one type of octa core Won't tell us about big.LITTLE octa cores since the MT6795 is 8x A53 Your study would be nice to test out how well performance scales And it would test if an octa core performs better than a quad core or dual core with the same cores But it won't test if octa cores perform better than dual or quad cores with different cores Ideally, I'd like him to do your study, but with all the different SoCs I named above

posted on Sep 04, 2015, 6:30 AM 0

21. TylerGrunter

Posts: 1544; Member since: Feb 16, 2012

I finally found some study that looks serious on the matter:http://www.moorinsightsstrategy.com/wp-content/uploads/2015/07/Do-8-Cores-Really-Matter-in-Smartphones-Final-By-Moor-Insights-Strategy-updated-links.pdf Give it a look if you are interested, it also includes several 6 to 8 cores SoCs.

posted on Sep 04, 2015, 6:44 AM 0

22. vincelongman

Posts: 5745; Member since: Feb 10, 2013

That's a great study But it doesn't have anything at all on power consumption "We do acknowledge that having many smaller cores is one way to simplify power management, but these tests are not focused on power" The reason they started making 6+ core SoCs for mobiles was to try improve efficiency So its not surprising there's no performance boost Performance per a watt is what would be interesting

posted on Sep 04, 2015, 7:21 AM 0

10. tedkord

Posts: 17452; Member since: Jun 17, 2009

You're looking at it from a two dimensional point. I've been saying for some time now that your phone doesn't only run the one app you're using. It's got multiple threads going, and more cores will benefit whether the app you're using is single threaded, or optimized for 4.

posted on Sep 03, 2015, 10:39 AM 4

17. TylerGrunter

Posts: 1544; Member since: Feb 16, 2012

Sorry but I stick to what I said in my second comment: Can you PROVE that more cores benefit the user? The only serious study I know about it says did not reach a definitve conclusion about it. If you can't prove it then it is just your opinion and merely an assumption.

posted on Sep 04, 2015, 3:05 AM 0

6. Mreveryphone

Posts: 1841; Member since: Apr 22, 2014

#allcoresmatter Love it!

posted on Sep 03, 2015, 7:54 AM 2

11. marorun

Posts: 5029; Member since: Mar 30, 2015

So when they tested they used a single SOC and hardware deactivated core right? Because if they compared different soc then this test is bulls**t. Take a octocore soc. Do the test with all core. then with one cluster of 4 core then deactivate half thsts 4 core cluster and to the test. I know its may be almost impossible but this would be the only valid way to do it. Because different SOC = different architecture wich all affect end performance.

posted on Sep 03, 2015, 11:27 AM 2

24. alex3run

Posts: 715; Member since: May 18, 2014

Exynos 7420 is the most adequate choise for the testing. Snapdragons 810 have big cluster bias and overheating because of that, Mediatek is good but it acts the same as Exynos.