Tip Us!

All input fields are optional, fill out as much or as little as you'd like.

Why octo-core, and where will it end?

by on February 6, 2013 9:45 am
12
views7037
11
0
111

processor CPU graphic

Those who like to follow hardware will have no-doubt noticed the increasing prevalence of quad-core processors in top of the line products, and it looks like the trend is set to continue with the introduction of octo-core processors later this year.

We've already had our first snippets of information about Samsung's new Exynos 5 Octa processor, which will come in an eight-core variety. Then there's also Nvidia’s Tegra 4, which sports an overwhelming 72 core GPU on top of a five core CPU.

Is there any end in sight to this exponential growth in core numbers, and is it even the solution we really need to pump up the performance of our portable devices? Well, that's what I'm here to discuss.

Where did this trend originate from?

If you can recall back to the turn of the millennium and Intel's introduction of the Pentium 4 processor, you'll probably remember that even these decade old PCs were commonly clocked up above 3Ghz. Fast forward to today's modern PCs and you'll still find that even the most expensive Intel i7 doesn't venture beyond the 4Ghz wall without some tinkering and customized cooling solutions.

The problem is that processors hit the wall, in terms of clock speed, a decade ago, and in the search for ever faster computing solutions we've had to turn to multiple cores.

Samsung-Exynos-5-Dual

Looking more specifically at recent portable computing solutions, it's obviously important that the device is power efficient and that remains relatively cool. Higher clock speeds not only require more power, but also churn out more heat, which can't be cooled particularly well when you want your smartphone to only be 1cm thick.

This is where multi-core processors provide the perfect solution, a good balance of power and heat, but also allowing you bypass the performance problems associated with lower clock speeds. This is because each CPU core can run at lower frequency and voltage, compared with a single core chip, whilst maintaining performance.

What’s the maximum number of cores?

Now that we've established that running two cores side by side solves the problems of low performance in smartphone and tablet devices, similarly, you could argue that continuing to increase the number of cores for parallel processing would overcome any speed bottleneck. However this is where the multi-core solution starts to look less impressive, as there are diminishing returns as the number of cores increases.

Nvidia Tegra4's die, showing each of the 72 GPU and 5 CPU cores.

Nvidia Tegra 4′s die, showing each of the 72 GPU and 5 CPU cores.

Firstly there's the problem of wasted clock cycles, instructions need to finish at the same time otherwise you're processor will be wasting energy. In order to utilize every clock cycle on a quad or octo-core processor efficiently, a software developer needs to optimize their application to sync parallel processing, but this ads time and costs to software production. In some cases applications simply aren't very demanding anyway, so your cores are just going to sit there using battery.

One solution to this is to clock down the processor's speed when the cores aren't being utilized fully. But more powerful chips require a higher minimum amount of energy than slower processors, even whilst idling, which still causes wastage.

Then you also have to consider bandwidth constraints, in other words how much data can we feed to multiple processors at any one time. Once the bandwidth is reached you can't push any more data through, regardless of the number of cores available. This limits the real world performance improvements of multi-core over single core processor speeds, you'll find that other bits and pieces of hardware bottleneck the rest of your system.

So whilst theoretically you could add more and more cores, it doesn't necessarily allow you to push more data through at once. And when you're not using them, multiple cores can be a massive power drain.

The 8 core myth

Taking all of that technological jargon on board, it seems like there are certainly limitations when it comes to multi-core systems. But we've heard so many mixed messages from manufacturers over the years it's hard to tell who to believe. Intel for example talked of hundreds of processor cores back in 2011, and Nvidia seem certain that GPGPUs are the future. However, Qualcomm remains adamant that a few cores with higher clocks are the best solution. So who's right?

Well rather than listen to the cheap talk, when we actually look at processor specs you'll notice something quite interesting — 8 core ground breaking performance is actually a bit of a myth, at least for now.

Take Samsung's new Exynos 5 Octa processor for example. Although it's called an octo-core processor, all eight cores aren't working in tandem as you might think. Instead there are two sets of quad-cores, a set of powerful ARM Cortex A15′s and a set of low power ARM Cortex A7′s, only one of which can be used at any one time. An elegant solution for the power efficiency problem I mentioned earlier.

Whilst this allows the chip to be more economical, by switching to the faster power hungry cores only when needed, there's no performance benefit over a standard ARM Cortex A15 quad core chipset. Nvidia's Tegra processors work in a similar way, using a single low power core for low-intensity tasks and the quad-core only when needed.

ARM big.LITTLE technology shows the truth behind the Exynos 5 Octa. Side by side quad-cores, not a true octo-core.

ARM big.LITTLE technology shows the truth behind the Exynos 5 Octa. Side by side quad-cores, not a true octo-core.

As another example, take a look at the PC market for the past five years. The number of cores hasn't ventured above four in mainstream or high-end gaming products, despite the fact that a Q6600 is now over five years old. If you take a look at the only full 8 core product in the consumer market, the AMD FX-8 range, it actually performs poorly compared to better designed Intel quad-core processors. Even Intel's new Haswell chips aren't increasing the number of cores available; the PC market has clearly reached a plateau.

So why is it such a struggle to push past four cores, well it all lies in the nuts and bolts, it's down to transistors. Either way you look at it, a fast single-core or slower multi-core processor has to be made up of increasing numbers of transistors, and more transistors means more heat and more power consumption.

Problem 101 with handheld devices, there's only a limited amount of battery power and no room for cooling. So it's doubtful that a fast, real eight core chip is a viable solution on smartphones or tablets, at least in the immediate future.

Looking to the future

Don't misunderstand me though, the Exynos 5 Octa is an incredibly good idea for handheld devices, and is a decent indicator of what processors are likely to look like in the future.

It's doubtful that the next improvements in processor technology are going to come from simply slapping 16 cores together on one chip. Even Nvidia's boastful sounding 72 core GPU isn't ground breaking. High end graphics cards contain thousands of stream processors, or cores as Nvidia likes to call them.

ARM bigLITTLE

Instead it's more likely that we will see companies look for more efficient ways to run their processors, continuing to reduce the die size of their chips to save on heat, and looking to ingenious solutions, like interchangeable cores, in order to conserve power where they can.

Whilst we will certainly see more eight core processors in the future, they are unlikely to be multi-core chips in the current understanding.

To sum up

As strange as it seems ,we aren't actually venturing off into the realms of infinity any time soon, technology isn't quite moving forward as fast as the marketing hype would have you believe. We are probably going to need another technological breakthrough and a change in both hardware and software developer mindsets before we see huge numbers of real cores become the norm.

Even so, you can expect that “8-core” will be the marketing buzz word throughout 2013 and 2014 when it comes to smartphones and tablets. A better term would probably be “dual quad-core”, but I suppose that would be a little confusing.

As always, beware of product marketing, they love to let consumers misinterpret their words.

WRITER

Google+ E-mail

Long time Android user and supporter of anything open-source, Robert is a BSc (Hons) graduate with an enduring fascination for technology. Glued to his Galaxy S2 there’s very little tech news which slips past him unnoticed.

YOU MIGHT LIKE

Google-IO-2013 NVIDIA Tegra 4 Shield 2 1600 aa
Comments
  • Dean Vandaele

    they better optimize touchwiz some more, performance is already very good^^

  • Jusephe

    Entire post in one sentence:

    “Samsung is going to boost clock speed of exynos octa by 10% in summer. “

    • MasterMuffin

      Maybe you should read the article again…

    • http://www.garysims.co.uk garysims

      I don’t mention clock speed anywhere in the article, so no that isn’t a good summary of the post.

      • Zach hidden

        Thought it said something about changing clock frequency and clock speed?

  • IulianPeride

    I guess this could be something like:
    “samsung will work to reduce the lag in the gs4, 10% increased performance should be enough to make you have a fluid browser”

  • naresh kumar

    Is it normal..that the backside of the phone gets heated during moderate to heavy use?

    • IncCo

      yes

    • pio66

      Yes my htc one and i9500 both do

  • Noah

    That’s quite an architecture, ARM have always been good at managing power on their chips.

    Anyway, it’s not for Europe nor US, so I wonder where one could get a hold of those S4 models… and since I certainly won’t be alone, I wouldn’t be surprised if this ended up in messy stock consequences.

    • technology

      Come visit Israel :) We have plenty of S4s Exynos5 octa version :)

      • Noah

        Coming! ;)

        • SiskoNetwork

          Middle east (Qatar, UAE, Saudi Arabia, Palestine, Israel, Lebanon) and North Africa (Tunisia, Algeria etc) All have the Octa version.

    • Hikari0307

      Asian countries like Malaysia

  • http://www.facebook.com/kern.cheh Sim Kern Cheh

    I am very interested in the possibilities custom kernel developers will be coming out with the octa chip. Perhaps it is a matter of time before someone comes out with a scheduler that can force distribution of processes across all eight cores for some well known intensive tasks.

  • p3ngwin

    actually big.little has heterogeneous processing using ALL cores, called “big.little MP mode”.

    http://en.wikipedia.org/wiki/Big.LITTLE#big.LITTLE_MP

    in these videos you can see more than 4 cores running, a mixture of the A15 and A7 cores working simultaneously:

    http://www.youtube.com/watch?feature=player_detailpage&v=eJtmW5nzsiE#t=65s

    http://www.youtube.com/watch?v=bXvecQfwHOk

    Even Samsung’s own official presentation at MWC showed mixed-mod heterogeneous processing :: http://www.youtube.com/watch?v=U6UNODPHAHo&feature=player_detailpage&list=PL9yOcWwG6fF4HPl3Sucy4JD_GN0pI9tdj#t=59s

    • kascollet

      For the moment, the kernel can only manage 4 cores at a time, which can be a mix of A7 and A15 (for example 1 A7 and 3 A15).
      The forthcoming update will allow more than 4 cores to run at the same time : the most powerful combination possible (true Octa dissymmetrical computing).

      • p3ngwin

        would that be the Linux Kernel?

        i’m curious how the videos clearly show MP mode already working, are they not using the Linux Kernel in the videos

        • http://www.garysims.co.uk garysims

          How do you know what kernel they are using in the videos?

          But to answer your questions Android uses a variant of the Linux kernel and the patches from Linaro for the migration mode and for the MP mode are for the Linux kernel.

          • p3ngwin

            “How do you know what kernel they are using in the videos?”

            i don’t, that’s why i asked.

            regardless, the videos clearly show both A7 and A15 cores being used.

            the “number” of cores used is irrelevant, as the definition of big.little MP mode isn’t defined by a number of cores, but in fact by the “type” of cores used.

          • http://www.garysims.co.uk garysims

            Yes the video does show A7 and A15 cores being used, but in pairs where one A7 is shut down in favor of its partner A15. That isn’t HMP.

    • http://www.garysims.co.uk garysims

      I have looked at those videos and none of them show HMP or big.LITTLE MP as ARM call it.

      • p3ngwin

        ARM describes using BOTH A7 and A15 of cores simultaneously, which is precisely what the video shows.

        i don’t understand what you read as ARM’s definition, or what you see differently in the videos.

        • http://www.garysims.co.uk garysims

          As I mention above, using a combination of A7 and A15 cores where an A7 is paired with an A15 and one or the other is used (depending on the load) isn’t HMP. In this context ‘simultaneously’ means that all the A7 cores can be used along with all of the A15 with each core running independently and receiving tasks from the scheduler which knows that some of the cores are less powerful than others.

    • Zach hidden

      So is this true AND the article true, but NOT the video? Because the article says more than 4 cores, this guy says more than 4 cores but the video only has 4 cores working at once

      • p3ngwin

        the article says:

        “the Exynos Octa uses the A7 cores by default and then when things get busy the A15 cores are activated while the A7 cores are shut down. This means that the Exynos Octa only uses a maximum of four cores at once,”

        So the article states *either* the A7 *or* the A15 cores can be used.

        Meaning currently the chip can NOT use BOTH A7 and A15 simultaneously, which is why i say it already CAN, as shown clearly in the videos i shared.

        I’m not disputing the current limit of 4 cores used, but the assertion A7 and A15 cores aren’t currently already being used simultaneously.

        • http://www.garysims.co.uk garysims

          p3ngwin, Now I see where the confusion is… I am partly to blame as I was trying to explain a complex issue and I left some room (unintentionally) for ambiguity. Yes of course the Exynos Octa can use a combination of A7 cores and A15, the switching is on a per core basis.

          As it says later in the article, “The switching from one core to another isn’t actually handled by the chip itself, it is done in software. In this case the Linux kernel which is found at the heart of Android.”

          The big.LITTLE desihn allows for the A7 and A15 cores to be grouped together, say in pairs, and ONE A7 core is shut down in favor of an A15 core which is started in its place. In that sense the processor is using A7and A15 at the same time.

          However the point of HMP is that all eight cores are used at the same time and the scheduler is clever enough to know that you can put the gmail sync task on an A7 but complex apps go on the A15.

          • p3ngwin

            yes, you did leave much of the important detail out of the article when making such claims. as for what defines HMP, it’s not the amount of cores used, as that’s more related to the amount of work needed to be done re: threads, etc.

            HMP is defined as simply using different processors, and in this case using ANY amount of both a7′s and A15′s qualifies as Big.Little MP (HMP).

            E.G. Nvidia’s 4+1 architecture is not HMP is any way because the processor has no way to make use of the different cores simultaneously. compare this to the Samsung processor, which *does* have the ability to simultaneously use the different A7 and A15 cores, and hence qualifies as Big.Little MP (HMP) regardless of how many of each core is used.

            here’s another video from ARM using the Samsung chip demonstrating HMP: http://www.cnx-software.com/2013/02/16/arm-big-little-processing-demo-hmp-on-arm-tc2-test-chip/

          • http://www.garysims.co.uk garysims

            p3ngwin,

            With regards to this statement: “HMP is defined as simply using different processors, and in this case using ANY amount of both a7′s and A15′s qualifies as Big.Little MP (HMP).” I am sorry but you are completely wrong. I have spoken to ARM personally about this and I don’t think you have understood the principles.

            HMP (rather than SMP) means that the scheduler needs to know that the environment is heterogeneous (meaning that the processing unit, in this the cores, don’t have equal processing power).

            In the current configuration that Samsung uses (and others) the scheduler only knows about fours cores (virtual cores) and using the techniques employed by the CPU frequency driver the different cores are switched from A7 to A15 (or back again) according the current load. The scheduler thinks that all four cores are A15 and the power is saved by flipping between the two. I know this as I have downloaded the source code from Samsung and read it!

            On a current dual or quad core CPU the frequency changes according to the work load, in the big.LITTLE migration mode the same thing happens but when the frequency gets to a high threshold the core is flipped and when it gets low it is flipped again.

            That isn’t HMP, it is SMP with some clever swapping of cores to save power. Each core shares the same L2 cache and there is some clever hardware to make sure that pipelines of both cores are identical so that the context switch is quick.

            HMP is different.

            Here is a quote from ARM’s whitepaper on the subject:

            —–
            Execution modes in a big.LITTLE system

            1. big.LITTLE migration model
            2. big.LITTLE MP model

            As the name suggests, the migration model enables the capture and restoration of software context from one processor type to another. In CPU migration, each CPU in a cluster is paired with its counterpart on the other cluster and the software context is migrated opportunistically between clusters on a per-CPU basis. If no CPU within a cluster is active, then the entire cluster and the associated L2 cache can be powered off.

            The MP model enables the software stack to be distributed across processors in both clusters. All CPUs may be in operation simultaneously offering peak system performance.
            —-

            Mode 1 – migration – is seen as an SMP system to the scheduler. Mode 2 – MP – is seen as a HMP system to the scheduler. The current kernel in the Galaxy S4 works in mode 1. Linaro have code out for Mode 2.

          • http://www.garysims.co.uk garysims

            Another thing… Re: “compare this to the Samsung processor, which *does* have the ability to simultaneously use the different A7 and A15 cores, and hence qualifies as Big.Little MP (HMP) regardless of how many of each core is used.”

            Yes, the Exynos 5 Octa does have the ability to run in HMP mode, that is the whole point of the post! But it needs to be used in that way and the software (in this case the Linux kernel) needs to be able to use it in that way. That is why the idea expressed in the article is that Samsung *could* release an update which switches the processor from big.LITTLE migration mode to big.LITTLE MP mode, from SMP to HMP.

          • p3ngwin

            ok, i’ll concede i didn’t understand fully that ARM’s definition of “Heterogeneous Multi-Processing” is slightly different from the generic definition used in most other processors.

            ARM have “Cluster Migration”, “Core Migration” and then what they deem as “HMP”.

            As far as i’m concerned “Core migration” *is* heterogeneous processing already, but if ARM want to sub-divide such generic definitions and deem only the latter as HMP then so be it.

            It just would be more sensible NOT to use existing generic terms for a specific technique.

            It’s almost like Americans renaming “Petrol” as “Gas” :)

          • Anirudh

            They wont allow me to double like this comment.

  • vampyren

    Normally I would be pissed but even my s3 was a monster and s4 with s600 is almost 3 times faster so I don’t care much about more performance really. Samsung should do more software optimization for better battery and bug fixes on both phones. I get my s4 next week so I don’t know yet how well everything works but I was very happy with s3.

    • Zach hidden

      Dude I’m still happy with my s2 skyrocket(the LTE model of the s2) it had problems with the 4.0
      . Something update but the 4.1.2 is real nice, because of its speed

      • vampyren

        Yes JB made the phone much smoother. On the otherhand the S2 was a great phone but i couldnt stand the screen resolution coming from iphone so i kept using my iphone but once S3 came thats when i started to use Android seriously. Give the new one a try and you wont be disappointed, the screen is amazing.

        • Sujay Vilash

          HTC One’s screen is even better. And the phone, well, brilliant. Now if someone could just take the best of the S4 and combined it with the best of the One, they would have an absolute market leader on their hands.

          • vampyren

            I like AMOLED more , somehow colors look nicer and i like the blackness. Anyway at this point both screens are pretty awesome but for me as a whole i prefer touchwiz and the nerdy little features samsung offers. Plus i like the SD+battery + the material more on S4, i’m not very fond of aluminum personally.

          • MasterMuffin

            Why not best of everything right away :D

  • Dave Weinstein

    I’m ALREADY pissed about their processor “bait and switch”. The worst possible thing they could do would be to withhold optimization software in the name of appeasing the people they’re already cheated!

    • freedomspopular

      Dude, get a grip. They didn’t have the Octa ready in time, on top of the S4′s overwhelming demand..

  • Bone

    Do I always miss the announcement of AndroidAuthority giveaway winners or is it done in private?

  • PrincessR

    I just got my s4 here in the US with At&T. Experienceing massive battery drain. I dont think thr processor had anything to do with it ive switched off a lot of the features and even reduced the brightness but the phone does not even last a day. I hope Samsung is aware of this and not issuing the update only to regions using the exynos processor

    • kascollet

      That’s weird. Battery benchmarks seem pretty good.
      Do you use auto-brightness outdoor ?

  • http://ARMdevices.net/ Charbax

    Linaro IKS (current big.LITTLE solution): http://www.youtube.com/watch?v=1Ag-sLulFfY

    Linaro MP (future big.LITTLE solution): http://www.youtube.com/watch?v=Svc4pEvTmCs

    I filmed many more videos with these guys that are right now working on optimizing these solutions.

  • Chris Tyler

    You note that Linaro’s code… “is available to ARM’s partners including Samsung”. Actually, Linaro is an open source engineering company, and therefore the code available to everyone. (Go hack!)

  • Adam Outler

    People are complaining that the Exynos based GS4 is laggy. This is simply what they should have delivered in the first place.

  • Anonymousfella

    Please just focus on software optimization(decreasing the time to open simple and basic apps like camera, gallery) and increasing battery life…small things but they are the ones which actually count and make a difference in user experience!
    That a15 quad core will be powerful enough to run high end apps for the next 2years and such an update can always be pushed later in th product cycle…remove even the slightest hint of lag in the UI and gentlemen, we have a masterpiece on our hands….

  • Zach hidden

    Dude when I get the s4 in October ima buy a ballistic case ( OK protection and price not the best though) and an extended battery, take the plastic part of the case and melt and sand a perfect cutout to fit the phone+ extended battery. BOOM s4,eight cores, 16hr battery :-) and the s4 must be yellow

    • Zach hidden

      Now I gotta find an s4, eight core model yellow that has 4g LTE and works LTE with AT&T, to the eBay’s!!!

  • Anirudh

    Activating all 8 core requires to be thought deeply in terms of hardware also. Changes pertaining to heat dissipation. These SoC were not built (thermal design) to handle heat off from 8 core simaltaneously. So i believe HMP is one good thing but activating them in already-in-market devices is one big thing.

Would you like to view our mobile friendly site? Try it out