By: David Kanter (dkanter.delete@this.realworldtech.com), May 18, 2013 7:02 am
Room: Moderated Discussions
Exophase (exophase.delete@this.gmail.com) on May 17, 2013 10:41 pm wrote:
> David Kanter (dkanter.delete@this.realworldtech.com) on May 17, 2013 9:29 pm wrote:
> > Android applications don't really use GCC as I understand it. They are using Dalvik.
>
> Android has NDK which uses GCC. It was stuck with 4.4.x for a long time which held it back in
> ARM performance (code gen was a ton worse back then). It's at 4.6.x now which is better.
>
> A lot of things do use NDK, and some of it uses NEON intrinsics or even ASM.
> Think about all the software that's available on both iOS and Android. Writing
> a common core in something like C++ is usually a lot less work than
>
> A lot of libraries and middleware use NDK too.
>
> And maybe most relevant, most of the non-Javascript benchmarks (although unfortunately Javascript
> still make up a big chunk). The Javascript benchmarks are not that representative of anything
> outside of web browser performance and even that can be questionable, but as far as a lot of people
> are considered phones and tablets are just web browsing devices so maybe that's fair.
I think there are a few typos in here, but I get your general point. If you want a proxy for NDK performance, I would argue for SPECcpu2K. Or actual NDK apps.
> > So you're comparing two entirely different software stacks and trying to draw
> > conclusions. Moreover, my understanding is that Moorestown actually matches
> > the A9 and A15 on quite a few benchmarks (based on discussions with Anand).
> >
>
> What Moorestown hardware was even tested? Or are you just calling it all the same?
>
> I've seen one Atom core running two threads do well against one A15 core,
> but I haven't seen any single threaded scenarios where the Atom wins.
Sorry, I meant Medfield. And yes, I wouldn't ever expect comparable single threaded performance.
I don't test mobile very much, largely because it's a pain in the ass and my reviews get far less traffic than microarchitectural articles. However, in talking with Anand he was very clear that Moorestown is comparable or better than A9-based designs and sometimes can match A15.
David
> David Kanter (dkanter.delete@this.realworldtech.com) on May 17, 2013 9:29 pm wrote:
> > Android applications don't really use GCC as I understand it. They are using Dalvik.
>
> Android has NDK which uses GCC. It was stuck with 4.4.x for a long time which held it back in
> ARM performance (code gen was a ton worse back then). It's at 4.6.x now which is better.
>
> A lot of things do use NDK, and some of it uses NEON intrinsics or even ASM.
> Think about all the software that's available on both iOS and Android. Writing
> a common core in something like C++ is usually a lot less work than
>
> A lot of libraries and middleware use NDK too.
>
> And maybe most relevant, most of the non-Javascript benchmarks (although unfortunately Javascript
> still make up a big chunk). The Javascript benchmarks are not that representative of anything
> outside of web browser performance and even that can be questionable, but as far as a lot of people
> are considered phones and tablets are just web browsing devices so maybe that's fair.
I think there are a few typos in here, but I get your general point. If you want a proxy for NDK performance, I would argue for SPECcpu2K. Or actual NDK apps.
> > So you're comparing two entirely different software stacks and trying to draw
> > conclusions. Moreover, my understanding is that Moorestown actually matches
> > the A9 and A15 on quite a few benchmarks (based on discussions with Anand).
> >
>
> What Moorestown hardware was even tested? Or are you just calling it all the same?
>
> I've seen one Atom core running two threads do well against one A15 core,
> but I haven't seen any single threaded scenarios where the Atom wins.
Sorry, I meant Medfield. And yes, I wouldn't ever expect comparable single threaded performance.
I don't test mobile very much, largely because it's a pain in the ass and my reviews get far less traffic than microarchitectural articles. However, in talking with Anand he was very clear that Moorestown is comparable or better than A9-based designs and sometimes can match A15.
David