Back to Article

  • mkaibear - Tuesday, May 12, 2015 - link

    It's getting a bit like the gillete model now.

    Two clusters... no wait, 3 clusters! wait! 4 clusters! Low, middle, high, eXtreme Performance! wait! 5 clusters! With a lubricating strip!
  • Raniz - Tuesday, May 12, 2015 - link

    Except that here it may actually make sense. Reply
  • close - Tuesday, May 12, 2015 - link

    The main target isn't battery life, it's the marketing position. In a world where it's harder and harder to differentiate your product slapping on 2 more cores will put you on the radar. This is also true for resolution where you see ludicrously high resolutions on tiny panels (already talking about 4K).
    Plus it's a lot easier to slap on cores and some more RAM then it is to optimize your software and not have a "fart generator" type app that needs 50+MB of RAM or who knows how much CPU.

    Apple must be doing something right with the iPhone 6 if they can fight a much beefier Galaxy S6 (with a huge difference in terms of cores, frequency and RAM) and still be competitive performance and battery wise. And I don't think it's "magic", it's just good hardware and software engineering.
  • menting - Tuesday, May 12, 2015 - link

    Apple can optimize their CPU architecture to their needs only, and that's why they can compete well even with spec that is less on paper. Reply
  • Frenetic Pony - Tuesday, May 12, 2015 - link

    Apple's needs for the Iphone are the exact same as anyone else's needs for a smartphone. They're just much better at CPU architecture than ARM. Which is why I'm more interested in seeing Qualcomm's newest Snapdragon later this year, as actual Core design seems far and away more important than any amount of clever SOC design, as Apple's 2 cores fits all and still beats 10 different cores strategy shows. Reply
  • dccafe - Wednesday, May 13, 2015 - link

    Just so you know, Apple Iphones uses ARM architecture processors.
    The only difference are the add-on features, they are designed to better fit their software design.
  • rocketbuddha - Wednesday, May 13, 2015 - link

    FP is correct.
    ARM is a ISA. The company also develops cores running its architecture under the Cortex moniker.

    In general, QCOM, Marvel, Apple are companies that license the ISA, but design their own cores around it. NVIDIA tried the same with Denver, Not unlike AMD & INTC in the x86 world. Different designs running same/similar x64 ISA.

    But other manufacturers license the ARM Cortex cores and its CPU system architecture, add other components to it and make a chip/SOC at different manufacturing nodes. Samsung, MediaTek, LG, Rockchip, NVIDIA, Huawei are in this category.

    In the last generation QCOM hit the jackpot with the Krait core that was better at performance and power against the ARM Cortex A9 and better power/perf/die size advantage vs. ARM A15+A8 combination. With the modem related integration it basically whalloped competition.

    Apple dropped their first bomb by designing their own Swift core with ARM v8 compatibility before even ARM was ready with its own Cortex implementation of v8 ISA.

    QCOM pooh-poohed and missed the 64 bit bus very badly. But it had to rush the first generation 810, 610, 410, 808 in place by actually implementing Cortex A57 + Cortex A53 cores as Samsung and MDTK were expected to be ready before their custom QCOM's ARM v8 core was ready.
  • WinterCharm - Wednesday, May 13, 2015 - link

    Except apple's ARM cores are custom designed, since they have the license to take liberties with the architecture. Reply
  • babadivad - Saturday, May 16, 2015 - link

    Apple's cores are 100% custom. They don't take reference designs from ARMs Cortex CPU and modify them. They build them from scratch without any of ARM's input on design. But they did license the ARM ISA so their CPUs run ARM instructions [like Intel and AMD CPUs run x86]. Reply
  • Jef.Holt - Thursday, August 13, 2015 - link

    Apple has complete control of the entire system, unlike any other manufacture. Samsung comes closest with their SoC and building their own phones. But Apple has the CPU, phone and OS all under their control. This means Apple can create opcodes in silicon specifically to optimize iOS performance.
    So when you control the Integrated Development Environment (IDE), the hardware SOC (A9), the OS (iOS) and the additional components to make a turn key product, you can optimize every component to maximize the entire product performance. From the IDE perspective there is Metal and other tweaks under the hood that create tighter more efficient code. This code is so efficient that both Google and Microsoft have porting products out that take advantage of that same tight code to both improve performance of there phones and to quicken the porting process.
    SoC the A7, A8 and A9. All of these chips have CPU, GPU and IPU, just to name a few, that can or do have Apple specific code to optimize the performance and efficiency that you must have total control to get. We can see this in the real world usage more than the Bench Marks because code is more often optimized for Apps than for the Bench Marks.
    Other chips used. Apple can test multiple chips for efficiency in their specific designs to find the most efficient combinations, this maximizes the battery and allows for smaller lighter devices. Apple can also use features of the support chips much more efficiently because it can write the usage directly into the IDE. Other manufactures cannot do this because they use the monolithic Android OS that is written with the idea of all SoCs and all support chips.
    All of these optimizations add up to added performance that we can see.
  • mortimerr - Tuesday, May 12, 2015 - link

    I've been wondering alot about this. I love the Android ecosystem and I'm not sure I could leave. I've tried understanding what exactly allows Apple to achieve their performance.
    Apple people just say "The hardware and software are designed by the same people" but that doesn't explain how. How does the architecture of the A8 differ from the 810 for instance? Is there literature available on this?
  • dmunsie - Tuesday, May 12, 2015 - link

    I think it's a matter of Apple only adding things to their SoC that will give them the biggest return on investment. They have the ability to profile their entire OS stack and all of the applications on the store and find out exactly where the most benefit would come from.

    On the Android side, you have Google providing the core OS, OEMs customizing the OS and adding their own enhancements and the SoC manufacturers trying to sell to those OEMs. I'm sure that each company is doing their best to make the best product they can on their own, but they have little to zero impact on what everyone else is working on. So they make changes where they can which may have less of an impact than the things that Apple can change in their entire stack.
  • phoenix_rizzen - Tuesday, May 12, 2015 - link

    Anandtech does a good job of diving into all the technical details of the Apple SoCs when they are released. They do multiple articles on each SoC. Just read through those, and then compare them to the articles they write on the Qualcomm/Samsung SoCs. Reply
  • jjj - Tuesday, May 12, 2015 - link

    Most of that is just not true.
    Apple's core is bigger and in single core vs single core, Apple is faster now. Overall with 2 cores for Apple vs 4xA53, Apple loses hard. Apple's core is not all that great,they just aim for 2 cores instead of 4 in a certain power limit.. We should see bigger ARM cores soon ,that will likely land in 2 cores configurations in phones.
    Until then, the A72 at 2.5GHz is a lot faster than what Apple has now on a similar process.
  • NetMage - Tuesday, May 12, 2015 - link

    Except, I'm not sure that is what matters in a final product - Apple's phones are competitive in battery life and performance to Android phones and they do that with two 64-bit cores. Reply
  • jospoortvliet - Friday, May 15, 2015 - link

    I'm sure the Android phones could be competitive in battery life and performance with two cores, but people wouldn't buy them as they think more cores is better... :( Reply
  • WinterCharm - Wednesday, May 13, 2015 - link

    Anandtech does a good job of looking at exactly what makes the A8 so superior. What it comes down to is that Apple designed their mobile chips architecturally more similar to Intel Desktop CPU's, than they did to any ARM processor out there.

    Read the first 3 pages of this iPhone 6 review:

    It's a very detailed account of what the A8 chip is, and how it works, and what architectural changes and improvements apple has made. One of the most incredible features is carried over from that A7 chip in the 5s - a 192 entry reorder buffer... exactly the same as Intel's Haswell chips have.

    Apple calls their architecture "Cyclone" and claimed in that keynote 2 years ago that it was "desktop" class. They weren't kidding.
  • tuxRoller - Tuesday, May 12, 2015 - link

    ... And this is why you don't judge a cpu based purely on GHz, core count, memory technology, or any other single number than IPC/W, or something similar.
    The apple cores are a great deal more complex than any other native arm core. They're HUGE. They're wide. They've an excellent bpu.
    Still, with all their advantages the iPhone don't really enjoy much, if any, perf/W over the best android implementations.
  • nandnandnand - Tuesday, May 12, 2015 - link Reply
  • BillBear - Tuesday, May 12, 2015 - link

    This was absolutely the first thing that sprang to mind when I saw this story. Reply
  • jay5 - Tuesday, May 12, 2015 - link

    Think the Snapdragon 808 in the table should be listed as having 2x A57s not A72s Reply
  • Ian Cutress - Tuesday, May 12, 2015 - link

    Yup, a mistake in our copy and paste. Updated :) Reply
  • rocketbuddha - Tuesday, May 12, 2015 - link

    Hi Ian,
    Could you also correct/clarify the below

    MT6795 - 4xA53 2.2GHz + 4xA53 2.2GHz.
    I think that the 2nd cluster should be like 1.3-1.5 GHz...
  • JoshHo - Tuesday, May 12, 2015 - link

    The X10 SoC isn't quite a conventional bL SoC. Both clusters are clocked identically and treated identically. Reply
  • MrSpadge - Tuesday, May 12, 2015 - link

    Given the tiny size of A53 cores this could really make sense. And under heavy multi-threaded loads (image & video processing) one could imagine 8xA53 to be more efficient than using fewer big cores. I don't know if they can be used together, though.

    There could also be further synergy between the clusters by sharing a last-level cache. This could probably reduce average switching latency and provide more cache space (or a lower die size). Caches are also built for certain power & performance points, though. so sharing the cache between the low and high cluster would hurt power efficiency and/or performance.
  • Kracer - Tuesday, May 12, 2015 - link

    Why all this complexity to gain CPU performance on mobile? That is not to say to stop improving performance, but the experience is rarely ever constrained by CPU perf. GPU perf, battery life and storage performance have a tangible effect in the experience of a smartphone.
    A pair of very low-power A53s and a pair of blazing fast A72s(similar to the cluster of this SOC) are plenty. This setup can rush to idle to give a responsive experience and be very low-power.

    If one is doing a sustained workload on mobile that is not offloaded to a dedicated ASIC(video/audio playback/capture) and are not plugged in they are using a smartphone wrong.
  • psychobriggsy - Tuesday, May 12, 2015 - link

    Because it isn't that complex for the SoC designer to put the IP blocks together (assuming their custom interconnect is a modified ARM design), the A53 cores are tiny so you might as well use four (and maybe licensing tiers encourage using a quad-core unit rather than a dual-core unit), and including the low power co-processor reduces integration cost.

    The A72s do sound like a benchmarketing mode to me, but sometimes race to sleep really does work (with timer coalescing and other OS level features supporting it).
  • jjj - Tuesday, May 12, 2015 - link

    If they do get 30% power savings, or even 20%,then it's worth using an extra 4.5mm2 (more or less) for the additional quad cluster. Any clue on cache sizes, maybe they save some die there?
    This vs SD620 shouldn't really be any contest given the 28- 39% higher clocks and likely more advanced process. Still no clue what GPU the SD620 will have but if it's aimed at upper midrange they got to keep the die size in check even if 28nm might be not that expensive anymore.So i assume they'll have to make some cuts to keep the die size at 80-100mm2 to sell it at 20-25$, if it's on 28nm.
    On shipping timing, Mediatek a couple of weeks ago during their results call said it's shipping to device makers in Q4 ,was kinda hoping we might see 1-2 phones late in Q4.
  • jjj - Tuesday, May 12, 2015 - link

    "MediaTek is proud to present its first CDMA2000 compatible integrated modem with the X20"

    The MT6735 and MT6753 have CDMA2000 and are in devices now so unless somehow those have it on a discrete chip, the HX20 is not the first.
    What might be a first for MTK is carrier aggregation, unless some unannounced chip hits the market sooner.
  • dragosmp - Tuesday, May 12, 2015 - link

    "[...]only the M4 is in use in order to decrease battery life"

  • Ian Cutress - Tuesday, May 12, 2015 - link

    My fault - brain derp while editing. Changed! Reply
  • nandnandnand - Tuesday, May 12, 2015 - link

    "The 10 processors are arranged"

    Shouldn't this be "processor cores"?
  • Pissedoffyouth - Tuesday, May 12, 2015 - link

    Sounds great. Call me when they start to release sources, so I don't get burned again being stuck on an old version of Android. Reply
  • SeleniumGlow - Tuesday, May 12, 2015 - link

    While I'm very impressed by the Helio x20, it saddens me to think that all phones/tablets that will be launched in my country with this SoC, they'll be brought because of "10 core processor" marketing and not because of the improved battery life.

    Also, I see something interesting... about dual camera support (13+13 MP 30 fps). Does this mean that 3D enabled cameras using the SoC for processing would be possible? I'm tempted to look for a developer's board just to try out this dual camera thing... though I don't think there will ever be a developer's board...
  • nandnandnand - Tuesday, May 12, 2015 - link

    I doubt many consumers know or care how many cores are in their smartphone. AnandTech/tech press readers are not average. Reply
  • Pissedoffyouth - Tuesday, May 12, 2015 - link

    Uh, lots of people bragged about their quad core phones with the S3 came out Reply
  • close - Tuesday, May 12, 2015 - link

    The just have to know that 10 is better than 8. Nobody cares what they're talking about but they do know numbers (well... most of them do anyway). Reply
  • mortimerr - Tuesday, May 12, 2015 - link

    Sales of the G4 will be a good test to see if that's true. From the little I've seen (AnandTech doesn't have the review yet), I like it over the S6 (plastic body aside) but it has 6 vs 8 cores, and if it sales drastically less, that has to be looked at as a possible factor, regardless of the fact that it possibly runs better Reply
  • psychobriggsy - Tuesday, May 12, 2015 - link

    It's good to read about the innovations and different approaches in the SoC marketplace.

    I'm treating this as an octo-core A53 chip with a dual-core benchmark winner cluster added on top :-) The A53 might be a small core, but it's still very decent if you are upgrading from a two year old phone that probably had quad-A9s.

    What really matters is if the chip can run for a reasonable amount of time at specified CPU clocks, or if it throttles fast like the 810. It looks like it won't from what has been written.
  • watzupken - Tuesday, May 12, 2015 - link

    I wonder what is the practicality of having so many cores on a mobile SOC. I am not sure how much power saves are they looking at since if they are really into saving power, the number of cores should reduce not increase. This is purely marketing gimmick here from my opinion. Reply
  • mayankleoboy1 - Tuesday, May 12, 2015 - link

    I think a 2+2+2 would have been a better config. What processes/apps uses 4 cores? Most top out at 2 threads. Reply
  • phoenix_rizzen - Tuesday, May 12, 2015 - link

    Yeah, that would be an interesting setup to benchmark. Reply
  • testbug00 - Tuesday, May 12, 2015 - link

    And, MediaTek wins the prize for the stupidest phone SoC yet! Take out the moden and such and throw it into a cheap laptop or mini-desktop... Sounds perfectly fine.

    In a phone? Take away 2-3 of the A53 cores and all the A57 cores.
  • phoenix_rizzen - Wednesday, May 13, 2015 - link

    There aren't any Cortex-A57 cores in this SoC. Reply
  • ScottSoapbox - Tuesday, May 12, 2015 - link

    Why not just have 2 frequencies for a single A53 cluster? Reply
  • Marc GP - Wednesday, May 27, 2015 - link

    Because they work at different voltages, and the consumption is proportional to the square of the voltage so having a cluster at a lower voltage is a pretty big deal. Reply
  • ScottSoapbox - Tuesday, May 12, 2015 - link

    Typo: table lists X10 clusters at same frequency. Reply
  • Andrei Frumusanu - Tuesday, May 12, 2015 - link

    It's not a typo, that's how the SoC operates. Reply
  • phoenix_rizzen - Tuesday, May 12, 2015 - link

    There's also two instances of "choose" that should be "chose". Reply
  • azazel1024 - Tuesday, May 12, 2015 - link

    Seems to me that a smarter design would be a quad core design in a phone, 2 A72s and 2A53s. In a tablet a hexacore and or maybe octocore might make sense with 4 A72s and 4A53s, or 2 A72s and 4 A53s. Reply
  • djvita - Tuesday, May 12, 2015 - link

    I'll start caring about MediaTek when they provide documentation to their soc for developers of custom roms and update their firmware when a major Android version comes out. Allwinner does this, even providing ubuntu/fedora images to build on their mb's. Reply
  • serendip - Tuesday, May 12, 2015 - link

    I thought MediaTek were getting better regarding this but perhaps not. Looks like they'll keep pumping out chips and hoping people buy new phones every year instead of hacking older ones. Reply
  • Valis - Tuesday, May 12, 2015 - link

    Give me a SoC with 2x A72 (and 2x A53, if you must), plus the GPU in Tegra X1, that's be enough for my mobile devices. Reply
  • jeffkibuule - Tuesday, May 12, 2015 - link

    One has to wonder why Intel never pursued this multi-architecture, single-die design if it's so amazing. Reply
  • mkozakewich - Wednesday, May 13, 2015 - link

    Atoms have two sets of one or two cores on one die. So they kind of do, but only sort of. Reply
  • toyotabedzrock - Tuesday, May 12, 2015 - link

    This seems to be combating bad scheduling software with more hardware.
    Also the m4 doesn't have an MMU iirc which means it could be a victim of rigged mp3 files.
    Also the modem ap often is not controlled by the user which might make this a new NSA favorite phone given the mic will be always on and is connected to this ap that the user has no ability to monitor.
  • monstercameron - Tuesday, May 12, 2015 - link

    @Ian, what do you think about the probably fake snapdragon 818? Reply
  • tuxRoller - Tuesday, May 12, 2015 - link

    Where's the GPL licensed source for core pilot? Reply
  • iamkyle - Tuesday, May 12, 2015 - link


    Should be called MediocreTek.
  • iwod - Tuesday, May 12, 2015 - link

    Any reason why I cant use this on my NAS or other devices? I mean SoC are now so powerful and yet many computing appliance continued to ship with shitty CPU.

    The other question on my head, even Mediatek manage to get a decent modem in their SoC, yet apple cant?
  • tuxRoller - Wednesday, May 20, 2015 - link

    Because they aren't releasing the source code...contrary to the gpl. Reply
  • BillBear - Tuesday, May 12, 2015 - link

    I've asked for this before, but now that you guys have someone who knows how to selectively disable cores you really need to conduct a series of benchmarks that measure how well Android performs various common tasks with different numbers of cores enabled.

    We all know that desktop OS's have trouble using more than a couple of cores effectively unless you run specialized application software, for example Handbrake.

    I would like to know if the situation is the same on mobile.
  • Andrei Frumusanu - Wednesday, May 13, 2015 - link

    OK. Reply
  • zodiacfml - Wednesday, May 13, 2015 - link

    Brilliant. I just don't know if this will be cost effective for the company considering the additional die size. This reminds of AMD days where they are giving more cores versus a more advanced manufacturing process node of Intel. The more advanced process in the ARM world right now is Samsung which might be lower in power consumption and with better performance versus this Helio x20. Reply
  • Fidelator - Friday, May 15, 2015 - link

    Why is this better than having one cluster of A53s capable of going up to 2.0ghz, scaling down as required and one with the pair of A72s? Reply
  • Andrei Frumusanu - Friday, May 15, 2015 - link

    You can't scale the A53 as low in power in idle situations if it's optimized to reach those frequencies at the same time. Reply
  • Marc GP - Wednesday, May 27, 2015 - link

    Voltage. The low Cortex-A53 cluster is able to run at a lower voltage than the high Cortex-A53 cluster, and the consumption is proportional to the square of the voltage. Reply
  • Marc GP - Wednesday, May 27, 2015 - link

    The power savings come mainly from the voltage reduction, not the frequency reduction. Reply
  • Jimmy12 - Monday, April 25, 2016 - link

    Hong Kong Global Resources Electronic Expo, Oukitel X20 device

Log in

Don't have an account? Sign up now