Smartphones going overboard with # of cores?

[-Slacker-]

So, with Samsung's announcement on their 8 core exynos, and intel's 48 core mobile processor set for the middle of the decade, I have to wonder if RISC software scales better with more cores than what we usually get on windows...

http://www.engadget.com/2012/11/20/s...r-in-february/

http://www.engadget.com/2012/10/30/i...hones-tablets/


Can smartphones really take advantage of that many cores using real world applications, and not just the same couple of synthetic benchmarks we see in reviews?

What are some popular apps that run on more than 2 threads on android or ios?

[wlee15]

In the Samsung Exynos example only 1 group of quad core can be active at a time so when the device is heavily taxed the more powerful A15s are on while the A7s are put to sleep. Conversely when the device is being lightly used the A7s are turned on while the A15 are put to sleep.

[Haserath]

The 8 core exynos counts as a quad core due to the big.LITTLE approach.

One half fast quad/One half slow quad

Intel... I'm not sure how they plan to use that.

[ShintaiDK]

The 48 Intel chip is only PR. Just like the old 80 core chip. Its for research and showoff.

And as others said, the exynos is really only a quadcore.

[Cerb]

Originally Posted by -Slacker-

Can smartphones really take advantage of that many cores using real world applications, and not just the same couple of synthetic benchmarks we see in reviews?

They will be able to (not 48-core, but 4-8-core--and nobody has yet announced more than 4 for any commercial chip). The real concern is not having enough when the time comes. If everybody has quads by the time quads (and more RAM bandwidth, please) can be made really useful, it be a complete non-issue. Whoever is stuck with a dual-core of similar performance/core will be slow. And, it's not like they can just whip out a SoC in a few months.

FI, popular game X is made available for many mobile devices, and can use 3-4 threads. Do you want to be selling dual-cores for the months following that? Even people that don't need the extra cores, and don't even understand what the words mean, will be wary.

Also, there's the keeping up with the Joneses thing, for the time being. Tegra's got 4, so we all need 4, just because.

What are some popular apps that run on more than 2 threads on android or ios?

Any application that can use 2 threads can make you want to have 3 cores, because there are background applications to worry about. This is especially the case on Android. One more let's the OS and lighter background applications do their work without getting in the way, even when they might take a fair amount of CPU time.

[dma0991]

I'd wish they improve per core performance instead of trying to cram in more small/large cores on a single die. At least Apple got the right formula akin to Intel. Others deserve a face palm.

[Tuna-Fish]

Originally Posted by dma0991

I'd wish they improve per core performance instead of trying to cram in more small/large cores on a single die.

The cost in power of making a core 2x as fast is ~4x, while if you just double the core count you only increase power use by 2x.

It makes very, very much sense to stuff as many cores as you can possibly use in a smartphone. Especially since with modern clock gating, a core that is off uses no power.

At least Apple got the right formula akin to Intel. Others deserve a face palm.

The PA Semi team that Apple acquired to build their new chips is full of superstars. Among other key people, I believe it holds the highest concentration of the designers of the original AMD K8 anywhere. Their chips are not awesomely good because of their high-level design decisions (eg, 2 vs 4 cores), they are awesomely good because their low-level design was done by some of the very best people in the industry.

[ShintaiDK]

Originally Posted by Tuna-Fish

The cost in power of making a core 2x as fast is ~4x, while if you just double the core count you only increase power use by 2x.

It makes very, very much sense to stuff as many cores as you can possibly use in a smartphone. Especially since with modern clock gating, a core that is off uses no power.

Until you sit with a task that only runs on 1-2 cores and you wait...wait and wait.

Funny how 4 also seems to be the magic number in smartphones.

[dma0991]

Originally Posted by Tuna-Fish

The cost in power of making a core 2x as fast is ~4x, while if you just double the core count you only increase power use by 2x.

It makes very, very much sense to stuff as many cores as you can possibly use in a smartphone. Especially since with modern clock gating, a core that is off uses no power.

It doesn't have to be improved by a factor of 2. The A6 has put to shame most of other manufacturer's implementation, that's good enough proof that more cores isn't the way to go just yet. The problem with increasing core count is that the resources aren't being used to its fullest potential. Tegra has a single auxiliary core that works together with 4 more powerful cores but it didn't improve battery life significantly.

Originally Posted by Tuna-Fish

The PA Semi team that Apple acquired to build their new chips is full of superstars. Among other key people, I believe it holds the highest concentration of the designers of the original AMD K8 anywhere. Their chips are not awesomely good because of their high-level design decisions (eg, 2 vs 4 cores), they are awesomely good because their low-level design was done by some of the very best people in the industry.

I don't think it would've been as good if Apple went ahead with the core race. They know that their OS isn't up to task for more cores yet. Other manufacturers on the other hand uses them as a marketing gimmick.

[MarkLuvsCS]

Originally Posted by ShintaiDK

Until you sit with a task that only runs on 1-2 cores and you wait...wait and wait.

Funny how 4 also seems to be the magic number in smartphones.

Blame lazy programmers not revolutionizing programming models. I mean how hard can it be?

[ShintaiDK]

Originally Posted by MarkLuvsCS

Blame lazy programmers not revolutionizing programming models. I mean how hard can it be?

Haha.

Ye exactly. Same with unemployed people. Just start a business and live on it. How hard could it be! I mean really, here in my post I instantly fixed unemployment!

[Skurge]

Originally Posted by dma0991

It doesn't have to be improved by a factor of 2. The A6 has put to shame most of other manufacturer's implementation, that's good enough proof that more cores isn't the way to go just yet. The problem with increasing core count is that the resources aren't being used to its fullest potential. Tegra has a single auxiliary core that works together with 4 more powerful cores but it didn't improve battery life significantly.


I don't think it would've been as good if Apple went ahead with the core race. They know that their OS isn't up to task for more cores yet. Other manufacturers on the other hand uses them as a marketing gimmick.

Krait is about as fast as the A6 and Exynos 5/A15 is quite a bit faster. So I'm not sure where you are getting the idea that it put other chips to shame.

[dma0991]

Originally Posted by Skurge

Krait is about as fast as the A6 and Exynos 5/A15 is quite a bit faster. So I'm not sure where you are getting the idea that it put other chips to shame.

Reviews that I could find about it doesn't suggest that Krait is faster than than the A6. (iP5, Krait) (Sunspider : iP5, Krait) (Browsermark : iP5, Krait). Even a Krait powered HTC One X doesn't seem to show that it is better. Exynos 5 seems to be the only contender to the A6 but its only on the Chromebook for now, which suggests that it would take up more juice than a smartphone based Exynos 5 would for that extra performance.

The numbers suggests that the A6 is designed much closer to the Cortex A9 than the A15. If Apple were to design a part where it is closer to the A15 in the future, it'll be an interesting comparison against the future Exynos chips.

[beginner99]

I would rather have 2 faster cores than those 4 anemic ones.

[joshhedge]

Originally Posted by Skurge

Krait is about as fast as the A6 and Exynos 5/A15 is quite a bit faster. So I'm not sure where you are getting the idea that it put other chips to shame.

Bare in mind that Exynos 5/A15 is clocked higher than Apples A6.

[Haserath]

Nvm

[Skurge]

Originally Posted by dma0991

Reviews that I could find about it doesn't suggest that Krait is faster than than the A6. (iP5, Krait) (Sunspider : iP5, Krait) (Browsermark : iP5, Krait). Even a Krait powered HTC One X doesn't seem to show that it is better. Exynos 5 seems to be the only contender to the A6 but its only on the Chromebook for now, which suggests that it would take up more juice than a smartphone based Exynos 5 would for that extra performance.

The numbers suggests that the A6 is designed much closer to the Cortex A9 than the A15. If Apple were to design a part where it is closer to the A15 in the future, it'll be an interesting comparison against the future Exynos chips.

Those benches are heavily browser influenced. The Note 2 is very close to the iPhone 5 and beats it in some benches. And it is using Quad core A9. It makes minced meat of the optimus g that is quad core krait and all of that is down to software.

Sent from my Lumia 800 using Board Express

[grkM3]

Originally Posted by dma0991

Reviews that I could find about it doesn't suggest that Krait is faster than than the A6. (iP5, Krait) (Sunspider : iP5, Krait) (Browsermark : iP5, Krait). Even a Krait powered HTC One X doesn't seem to show that it is better. Exynos 5 seems to be the only contender to the A6 but its only on the Chromebook for now, which suggests that it would take up more juice than a smartphone based Exynos 5 would for that extra performance.

The numbers suggests that the A6 is designed much closer to the Cortex A9 than the A15. If Apple were to design a part where it is closer to the A15 in the future, it'll be an interesting comparison against the future Exynos chips.

Why are you using a java web browser to show cpu performance?The old samsung galaxy nexus with very out dated hardware blows the doors off the gs3 in sunspider because it uses a better web broswer from vanila android.

the nexus chromebook scores 600 in sunspider with a dual core exynos 5 cpu so we cant use a java web browser to show cpu perfomance.

[dma0991]

Originally Posted by Skurge

Those benches are heavily browser influenced. The Note 2 is very close to the iPhone 5 and beats it in some benches. And it is using Quad core A9. It makes minced meat of the optimus g that is quad core krait and all of that is down to software.

Sent from my Lumia 800 using Board Express

Regardless of its browser dependence, it still is a benchmark. You can have the most powerful hardware but without a decent software to compliment it and use its fullest potential, its gone to waste. Beats it in some but with Sunspider at 1015 and Browsermark at 174584, it isn't the mince meat that I was expecting.

Unless of course we're talking about raw compute performance, Android could do better in that. No one has a comprehensive collection of data for raw compute performance of smartphones so I can't use it as a comparison. Then again, referring to my statement above, raw compute performance isn't doing much good if the software isn't good enough to use it.

Originally Posted by grkM3

Why are you using a java web browser to show cpu performance?The old samsung galaxy nexus with very out dated hardware blows the doors off the gs3 in sunspider because it uses a better web broswer from vanila android.

the nexus chromebook scores 600 in sunspider with a dual core exynos 5 cpu so we cant use a java web browser to show cpu perfomance.

Why blame me? Blame on the fact that is the only data you could get from the reviewers. Sure, the score is dependent on the browser but as I said above, performance isn't everything if you can't use it properly. From this review, the Galaxy S3 is ahead of the Galaxy Nexus in Sunspider.

[grkM3]

Originally Posted by dma0991

Regardless of its browser dependence, it still is a benchmark. You can have the most powerful hardware but without a decent software to compliment it and use its fullest potential, its gone to waste. Beats it in some but with Sunspider at 1015 and Browsermark at 174584, it isn't the mince meat that I was expecting.

Unless of course we're talking about raw compute performance, Android could do better in that. No one has a comprehensive collection of data for raw compute performance of smartphones so I can't use it as a comparison. Then again, referring to my statement above, raw compute performance isn't doing much good if the software isn't good enough to use it.


Why blame me? Blame on the fact that is the only data you could get from the reviewers. Sure, the score is dependent on the browser but as I said above, performance isn't everything if you can't use it properly. From this review, the Galaxy S3 is ahead of the Galaxy Nexus in Sunspider.

Why even bother posting a browser bench when u can find benches on CPU numbers instead of how well a browser uses java?

There are other tests that show the power os the CPUs and I have both a nexus and a gs3 and my nexus scores 1100 bone stock and my gs3 scores 1500 on sun spider and I would never try to tell anyone the nexus has better hardware than the gs3

[soccerballtux]

Originally Posted by dma0991

Regardless of its browser dependence, it still is a benchmark. You can have the most powerful hardware but without a decent software to compliment it and use its fullest potential, its gone to waste. Beats it in some but with Sunspider at 1015 and Browsermark at 174584, it isn't the mince meat that I was expecting.

Unless of course we're talking about raw compute performance, Android could do better in that. No one has a comprehensive collection of data for raw compute performance of smartphones so I can't use it as a comparison. Then again, referring to my statement above, raw compute performance isn't doing much good if the software isn't good enough to use it.


Why blame me? Blame on the fact that is the only data you could get from the reviewers. Sure, the score is dependent on the browser but as I said above, performance isn't everything if you can't use it properly. From this review, the Galaxy S3 is ahead of the Galaxy Nexus in Sunspider.

no, browser benchmarks are completely worthless unless it's ~same browser version
calling Iphone browser vs Android browser a benchmark comparison is hilarious

[mikeymikec]

Originally Posted by ShintaiDK

The 48 Intel chip is only PR. Just like the old 80 core chip. Its for research and showoff.

Yeah, I remember a good few years ago they said something like that. Even reputable manufacturers engage in a bit of vapourware announcements.

[Skurge]

Originally Posted by dma0991

Regardless of its browser dependence, it still is a benchmark. You can have the most powerful hardware but without a decent software to compliment it and use its fullest potential, its gone to waste. Beats it in some but with Sunspider at 1015 and Browsermark at 174584, it isn't the mince meat that I was expecting.

Unless of course we're talking about raw compute performance, Android could do better in that. No one has a comprehensive collection of data for raw compute performance of smartphones so I can't use it as a comparison. Then again, referring to my statement above, raw compute performance isn't doing much good if the software isn't good enough to use it.


Why blame me? Blame on the fact that is the only data you could get from the reviewers. Sure, the score is dependent on the browser but as I said above, performance isn't everything if you can't use it properly. From this review, the Galaxy S3 is ahead of the Galaxy Nexus in Sunspider.

There is geekbench and linpack to name a few that show pure CPU performance. Android does fine. I point to the Note 2 again, when a quad core A9 hadily beats quad core Krait, it should show you that test is nothing more than a browser bench and a showcase for how manufacturer can optimise for their own chipsets.

[ShintaiDK]

Originally Posted by mikeymikec

Yeah, I remember a good few years ago they said something like that. Even reputable manufacturers engage in a bit of vapourware announcements.

Well its not vapourware, they actually have it there. Specially IBM is a number 1 in these PR announcements.

The problem is they dont always tell the entire story.

For example with IBM:
We have 250Ghz transistors! But we can only make a handful of them run at it at a time. And it will never be in a commercial product.


Or another IBM one:
We have 14nm ARM cores! But its a Cortex M0 test chip with less than 12000 transistors.

[dma0991]

Originally Posted by grkM3

Why even bother posting a browser bench when u can find benches on CPU numbers instead of how well a browser uses java?

There are other tests that show the power os the CPUs and I have both a nexus and a gs3 and my nexus scores 1100 bone stock and my gs3 scores 1500 on sun spider and I would never try to tell anyone the nexus has better hardware than the gs3

So you're saying that the reviewer is wrong? I'd like to see your numbers instead of a word of mouth.

Originally Posted by soccerballtux

no, browser benchmarks are completely worthless unless it's ~same browser version
calling Iphone browser vs Android browser a benchmark comparison is hilarious

It isn't completely worthless as that is what you do with the phone most of the time. Were you expecting to run Folding on the phone instead?

Originally Posted by Skurge

There is geekbench and linpack to name a few that show pure CPU performance. Android does fine. I point to the Note 2 again, when a quad core A9 hadily beats quad core Krait, it should show you that test is nothing more than a browser bench and a showcase for how manufacturer can optimise for their own chipsets.

Optimizing to meet the potential of the software is what you should do. Hardware and software goes hand in hand.