Mediatek Unveils Helio P20, P25, X30 mobile chipsets


MediaTek, the gargantuan Taiwanese chip manufacturer has unveiled its three newest chips, the Helio P20, the Helio P25, and the Helio X30 at an event just yesterday. Let’s have a look at what’s new! (MTK officially announce the info of X30, P25 chipset features in Vernee 2016 new products conference – Vernee product roadmap)

The Flagship MediaTek Helio X30

Those interested in the best that MediaTek has to offer will be interested in the company’s new offering in the Helio X30. Built on TSMC’s new 10nm process, the Helio X30 will keep the same tri-cluster setup as the previous Helio X20. Every cluster is clocked above 2.0GHz, from the four Cortex-A35 cores (2GHz), the four A53 cores (2.3GHz) and the two new A73 cores (2.8GHz).

The X30 will support up to 8GB of LPDDR4 RAM, 4K video playback and 4K recording at 30fps.

Gizchina News of the week


helio-p

The Helio P20 & the Helio P25

Mediatek’s Helio P10 chip has enjoyed heavy use amongst many chinese OEMs lately, equipping many popular smartphones such as the UMi Super and the Alcatel Flash Plus 2.

The Helio P20 and Helio P25 are both updates to the P10, enjoying a slightly higher CPU and GPU clock speed. Built on a 16nm process from TMSC, the P20 and P25 will both support LPDDR4x RAM, which will provide a much zippier app to app experience. The Helio P25 will also support dual cameras with optical zoom, where dual camera setups have been tried in phones like the Huawei P9 and the iPhone 7.

It will be interesting to see these new chipsets in new devices, and hopefully we will be able to see more of these chipsets in the market soon.

Disclaimer: We may be compensated by some of the companies whose products we talk about, but our articles and reviews are always our honest opinions. For more details, you can check out our editorial guidelines and learn about how we use affiliate links.

Previous Xiaomi Mi 5s Plus full specifications, photos & camera samples
Next Uhans A101 gets its first OTA update

108 Comments

  1. afghanPower
    September 27, 2016

    nm != mm
    (edit): Love your reviews dude!

    • September 27, 2016

      Edited, thanks.
      Also thanks! Hope to do many more!

  2. E8hffff
    September 27, 2016

    Mediatek is too slow to release enough SoC’s and or Chinese manufacturers want to keep making phones based on MK65xx and MK6735 series chips, so there’s no flair with this news at all. No to mention which Chinese phone maker has implementing 4K video or even slow motion since it was introduced in Mediatek SoCs. Ridiculous.

  3. realjjj
    September 27, 2016

    To add key features you miss.
    X30:
    GPU Imagination 7XTP MT4 @ 820MHz
    UFS 2.1 support , LTE Cat 10 , sensor hub, better audio, external 2×2 wifi ac
    They claim things like
    CPU 43% higher perf and 53% lower power vs X20
    100% higher memory BW and 50% lower power
    GPU 2.4 times the perf and 53% “power saving”
    Assuming lower power would be on a per same task basis.

    And P25 pushes the CPU clocks to 2.5GHz vs 2.3GHz for the P20.

    The X30 is a proper high end, we’ll see if hey got an X35 at 3GHz or X35 will be 2.8GHz.
    CPU perf should be great, GPU perf very good but not class leading. Given the config and clocks, the GPU should be at some 210GFLOPS so maybe some 15-20% bellow the latest iphones- Apple uses 6 cores at lower clocks.
    No major missing parts , it could be better here and there but it’s a big step forward , closing the gap with what others have in the very high end. That’s how it goes, step by step , they can’t take the lead with the first try.

    BTW, they haven’t unveiled the SoCs, these are leaks.

    • Yep
      September 27, 2016

      I have no idea how they could differentiate the X35 now SoC are so powerful. I hope it is true about a universal (ARM+X86) desktop UI operating system from Google that can start to make use of this power. Then 16GB RAM SoC with Quad core 3.5 GHz which are possible will make sense because they can be in convertibles and smartphones which dock into a monitor. Otherwise this is getting stupid.

      • realjjj
        September 27, 2016

        Chromebooks can run Android apps. That means that we’ll see Android apps aimed at desktop.
        So it’s normal for ChromeOS to become just a mode in Android , like Android Auto mode or VR mode.
        CPU perf is enough in phones now and can get much better. GPU perf has a long way to go.
        On 7nm in 2018, A73 could be clocked north of 3.5GHz in phones, especially if advanced packaging is used.

        A73 is a very small core, a quad cluster with 2MB L2 cache should be about 5mm2 on 10nm. I fully expect a much bigger core from ARM soon, likely targeting 7nm in 2018 and that would be a big perf boost.

        • Guaire
          September 27, 2016

          Ares’ performance should be closer to Apple’s custom cores. Also it would fit better for dual big core designs like Mediatek’s rather than quad big clusters IMO.

          Even if stock Ares can’t match with it, a customized by ARM one might be closer if their any customer would be interested in.

          http://cdn.wccftech.com/wp-content/uploads/2015/04/ARM10nm-635×322.jpg

          • realjjj
            September 27, 2016

            Yes , Ares or Ares like as that leak is old and some things might have changed since.

            The CPU perf is enough, the gain in normal usage is small at this point as you depend more on storage and connectivity perf.
            2x big core would be great for marketing but it’s a bit pointless and 4x A7x would be more efficient with no or little perf penalty in real world usage.By 2018 an A72 followup could offer 2x the perf of an A72 at 2.5GHz and that’s quite plenty.
            But marketing matters so the best we can hope for is 2x big cores that are almost always sleeping with 4xA7x at medium clocks doing the demanding work while small cores do 80% of the work.

        • Lazar Prodanovic
          September 27, 2016

          Well their is an need for the developer & ultra book class SoC & I think reference Quad core A73 on 10nm @ 2.8GHz consuming 1.200mW per core or under 4.5W in full usage on POP ARM lib just qualifies to brick Intel’s M offering considering both metrics performance & performance per W (as 2.8GHz A73 quad core is equivalent to the top M sku at 1.6GHz). Remains to implement the boost & turbo frequency implementations on ARM & to do it properly (in hardware).

    • Guaire
      September 27, 2016

      I guess X35 will have that clock speeds and actual X30 will be underclocked. Probably ~2.3GHz for the A73 and ~600MHz for the GPU.

      They did that dirty trick with the X20/25, MT6795/6795T and almost every other chip.

      • realjjj
        September 27, 2016

        X20 was supposed to be 2.3-2.5 GHz and yes they rebranded the top SKU and added a 2.1GHz SKU.

        With dual A73 on 10nm, reaching 3GHz should be doable. Huawei seems likely to have 2.6GHz quad A73 on 16ff this year. Remains to be seen if there is an X35 and if it’s 2.8GHz or 3GHz. Given the P25 surprise, i see 3GHz as more likely. They also need to beat others in single core so they’ll push as much as they can.

        Either way, X30 wouldn’t be much lower clocks, they got no reason to do that, business wise. At lower perf the X20/25 will be half the price so even the slowest X30 SKU needs to be well above it in perf. Not that the X35 branding has much of a relevance,, it’s just a branding.

        • Guaire
          September 27, 2016

          After Geekbench 4’s release, probably it won’t have a hard time at least matching Snapdragon 830’s single core performance assuming it will came with an overclocked Kryo.

          I think even a bit faster than 3GHz doable with 10nm A73.

          Strangely at both version of Geekbench, A72 doesn’t perform parallel to its higher clock speeds of both Kirin 955 and Helio X25 comparison to Snapdragon 650/652.

          • realjjj
            September 27, 2016

            Geekbench 3 total is heavily impacted by encryption and especially memory perf. Look at integer and FP scores.
            In 4 , both encryption and memory have a much lower impact but memory still distorts the total a bit. Not 100% that’s the main issue with scaling in GB 4 but has to be part of it

            I expect SD830 to use a slightly custom A73 and be branded as Kryo but with the note,” Built on ARM Cortex technology”. If they don’t have a much better Kryo, it just makes no sense to stick with it as it loses even vs A72 in perf and power.

    • Muhammad Yasir
      September 27, 2016

      look who just came outta their CAVE !

      *applause*

      • realjjj
        September 27, 2016

        What can i do if the new phones are boring as hell and there is nothing interesting.

        • Muhammad Yasir
          September 27, 2016

          ahaha … just messing with you buddy

          honestly missed ya !

          • realjjj
            September 27, 2016

            Been a bit more focused on electric cars and studying Tesla lately, it’s a bit less boring than phones with 1080p A53 and same looks as 2012.

            And the times are interesting there as oil is about to go away.

            • September 27, 2016

              I would definitely like to see more phones with Snapdragons in them, like the Zuk Z2

            • Muhammad Yasir
              September 27, 2016

              lel … whaddya kno !

              i’ve been gunning for an electric car for some time now !

              i WOULD LOVE TO OWN ONE IN PAKISTAN !

              any cheap ones coming to my region ?!

            • realjjj
              September 27, 2016

              The only electric car worth considering is Tesla Model 3 at this point but at 35k$ it is not cheap. hard to figure out what China has to offer here so not factoring those in.

              Will edit to add a bit more after the vid.

              https://www.youtube.com/watch?v=TIh2PjqfXkQ

            • Muhammad Yasir
              September 27, 2016

              nice description !

              im definitely interested in autonomous vehicles … but i just wanted a cheap car to drive to uni (1.5 hours by bus and less than 30 min if i drive by car)

            • realjjj
              September 27, 2016

              Other electric cars offer shitty value that’s why they are not worth it at this time.
              Niu has some new scooters, not sure if you’ve spotted those yet.

              Why don’t you get a pedelec https://en.wikipedia.org/wiki/Pedelec
              It’s a bike with an electric motor that boosts your pedaling. So you work a bit to stay in shape but a lot less effort and you got nice range. If you run out of juice ,no problem.
              Xiaomi’s pedelec is not great nor all that cheap but it’s on option i guess.

            • Muhammad Yasir
              September 28, 2016

              wow … wait.
              electric motor and pedals.

              won’t i just kill my feet if i pedal from home to uni and back ?!

              is this good for long distances …
              its over 13 Kilometers !!!

              im was looking at scooters but parents are not so open about the idea since the Traffic in Karachi is a bit dangerous for bike riders.

            • balcobomber25
              September 28, 2016

              Current estimates say there around 53 years of oil left on the planet. Some analysts say that number is closer to 100 as new oil fields and drilling tech (like fracking) is being explored.

              Electric cars are very interesting but they are decades away from being mainstream.

            • realjjj
              September 28, 2016

              That’s the opinion of clueless people with zero vision
              Tesla already offers better value than gas competitors today but that’s not the point as it is not about car sales anymore.
              As i mention bellow, car as a service with autonomous vehicles will force a very fast transition to electric. CaaS is much cheaper and more convenient.

              The day regulators ok full autonomy , oil hits a wall and there will be a catastrophic decline in demand over just a few years not decades. It all starts in 2018-2020. By 2025 even folks in developing nations could afford to pay 50$ per month to ride in a rather nice car.
              Oil is done.

            • balcobomber25
              September 29, 2016

              That;s the opinion of science and business. Unfortunately, in your little $50 a month dream, it is not practical or based on anything other than your dreams. Auto Industry experts estimate that by 2040 electric cars will have 35% market share. The problem is it will take another decade at least to convert existing engine manufacturing facilities. It will take a decade for companies to convert their current models. On top of all that is the costs associated with such a move, it won’t be cheap.
              It will cost Auto companies billions to completely stop selling gas powered cars. And then there is Big Oil who provides auto companies with funding. Oil isn’t going anywhere anytime soon, auto companies will use it till it runs dry. Neither of us will see a point in our lifetime when there are no gas powered cars on the road.

              While your opinion is a good one, it is not a practical one.

        • Marco Lancaster
          September 27, 2016

          It’s definitely boring lately..

    • Lazar Prodanovic
      September 27, 2016

      “The X30 is a proper high end, we’ll see if they got an X35 at 3GHz or X35 will be 2.8GHz.
      CPU perf should be great, GPU perf very good but not class leading.”
      Let’s disassemble those claims & see what we could really expect.
      The CPU; A73 vs A72= 10%+ performance per clock & 20% less power consumption, 10nm vs 16nm= around 25% less power usage. So per clock (A731+ 10nm)=110% performance and 60% power consumption. With a clock difference 2.3<2.8=22%, 110×122%=135% performance & up to 10% power saving (I am generous hire). Mid cluster will be spending the same amount of power & deliver 15% performance bump, small cluster won't save anything (52% power per clock, 33% higher clock) but it will be marginally faster (13%).
      RAM: doubling the frequency & going to DDR 4 (2xDDR 3 32 bit vs 4x DDR4 16 bit) won't really duble true output, additional latencies will hurt CPU performance & output will help GPU, estimated bandwidth gain is in the best bet margines of 60%.
      GPU; PowerVR Rogue 7XTP MT4 vs MALI T880 MP4 per clock around 40% more performance & equally more power consumption. 17% clock boost & the given 10nm advances brings us to the 42% bigger power consumption & adding a up to 20% performance gain from 60% bigger memory output brings us to (117%x120%=140.4) 140.4×1.4=196.56%. Not bad considering GPU performance but not to good considering power consumption either. Anyway this brings GPU to a category of being capable to support 2K resolution displays or capable for VR with FHD display resolution. Either way I am not happy, not that it would be much better if they did go With G71 aka T980 MP6 considering cost/performance/power consumption metrics (25% bigger & 30% faster wile consuming 20% more power [but if down clocked 30% so that performance is the same consumption would be 50% or compared to the 7XTP MT4 only 60% & comparable to the previous gen X20 I would call this a win, win solution]). If we include royalties & licensing costs probably the size difference of 25% cost for T980 MP6 (G71) would come to some 10% cost difference to manufacture & with ARM POP G71 libs to 0%.
      So what is to expect?
      Well considering X20 couldn't keep up max frequencies in demanding CPU or GPU tasks for a longer period of time & how the X30 doesn't bring any significant power savings compared to it just on CPU front things will mostly remain the same, on the other end GPU will use 42% more power so it will be much worser there. We also sow how core pilot is working bad in the X20 deca core so it will continue to be a sama story with X30. All in all a really bad design from my view point. Thing that passes me the most is that they still used dual M4 CPU cluster for sensors hub processing (so this is basically the 12 CPU quad cluster SoC) instead employing the same ISA V8 64 bit A35's as then they could be used as general purposes (all tasks) ones. Considering that the X30 SoC will cost around at least 35% more & that it will even more rise the final product cost as it will must employ 4x DDR 4 chips (instead of two DDR 3 ones 4GB) that will cost 2.5x (4GB) to 4x (8GB) times more I seriously doubt it will be commercially highly successful product.
      I am wondering how complicated it would be to make a turbo boost reserved clock frequencies on a ARM reference big cores like A73? As this would eliminate need for much things. Probably not to hard as Samsung is doing it (but not so good).

      • realjjj
        September 27, 2016

        Not really but will only address the DRAM cost claim because it’s by far the most ridiculous. DRAM sells on a per GB basis and you can fit a 2x2x2 config in a PoP, like Samsung does here https://www.chipworks.com/about-chipworks/overview/blog/samsung%E2%80%99s-galaxy-note-7-is-more-the-batteries

        • Lazar Prodanovic
          September 27, 2016

          Mediatek didn’t done POP of the X30 for memory nor they need to & that would actually only cost them more & make costumers unhappy (as they wouldn’t have options to pick RAM by them self [including amounts of it]). Samsung is only doing it for SSD’s, the M2 specification for the bord is small so they gain for nand dies that way. It remains to be seen on commercial SoC’s & it only makes sense for HBM2 (or newer) integration.
          I didn’t go with a claim that the 10nm manufacturing process is much pricier per transistor but did go with a claim that 40% bigger GPU counts in for around 22~25% more transistors on the whole SoC & certainly yields won’t be fantastic on the new & un perfected 10nm FF+ TSMC node so my assumption of 35% more cost is actually optimistic one.
          Samsungs 10nm FinFET node is much more mature (&still far from good yields) but also considerably worser than TSMC.

          • realjjj
            September 27, 2016

            Seems that you don’t know what PoP is. PoP means package on package, it’s been the default for a long time.

            The config you see in the note 7 is 8Gb x 8 for 4GB RAM, nothing special on the cost side and DDR4 prices are quite ok

            • Lazar Prodanovic
              September 27, 2016

              It’s a pre rooted lib of a functional block for a given lithography process. & if you implement it in the same chip (SoC) then it’s understud that you will have to go with the same manufacturing process, it also means that your SoC just become a Mammoth sized class one (considering how much RAM takes in transistor count especially if we speak about large amounts of it), it also means that the chances for wafer production mistakes are bigger by an order of magnitude. It costs more A: you will have more un functional dies, B: you can buy a RAM chips on older manufacturing processes cheaper.

              Igi Pop.

            • realjjj
              September 27, 2016

              lol we are talking PoP not ARM’s POP IP

              https://en.wikipedia.org/wiki/Package_on_package

            • Lazar Prodanovic
              September 27, 2016

              & MTK doesn’t produce RAM.
              So get back to the reality.
              In reality it costs more & makes a sense only with HBM (for now).

            • realjjj
              September 27, 2016

              You really don’t understand this conversation at all.

              You made a claim that is beyond absurd:
              “the final product cost as it will must employ 4x DDR 4 chips (instead of two DDR 3 ones 4GB) that will cost 2.5x (4GB) to 4x (8GB) times more Iseriously doubt it will be commercially highly successful product.”

              And now you are utterly lost. The DRAM is in a package using conventional wire-bond.You got a bunch of dies, a normal substrate and solder balls and, with PoP, you just stack the DRAM package over the SoC and that has been the default for a long time.

              DRAM sells on a per GB basis and, outside of extreme capacities on a per die basis , the size of the die doesn’t make much of a difference. DDR4 prices are very good compared to DDR3 , not a problem at all, even less so in 6months or more from now when X30 ships.If anything DDR3 would cost more.

            • Lazar Prodanovic
              September 27, 2016

              Ok it will still remain pricier than DDR3 & you will still need to use 4 packages at least for 4 controllers. It may not cost duble but it will cost more. DDR3 can’t cost more as its easier to manufacture & it still isn’t the obsolete product nor it will become that in the long time. Even DDR3 is marginally not usable for anything that is latency sensitive (not to mention real-time) like ray tracing, SQL search & cetera… DDR4 is much worse naturally so their is a real need for a switch to something else (like HBM). Again HBM is much bigger candidate for a PoP as the memory controller is on it (& will stay there).

            • realjjj
              September 27, 2016

              OMFG
              It’s a single package- apparently you don’t even know what a package is…

              It’s safe to assume that it is a standard 366 ball memory package, just like any other 64 bit LPDDR4 out there.

              Pricing for memory depends on supply and demand balance not on costs. DDR4 in desktop today has a less than 10% premium over DDR3 and folks are still ramping 20nm DDR4 but very soon DDR4 becomes cheaper. Today 1GB LPDDR4 is likely at about 4.5-5$.
              In mobile the DDR4 bandwidth is needed and DDR4 offers lower power per bandwidth and that’s critical when every mW matters.

              You still have no clue at all what PoP is. Anyway, HBM is not a solution fit for mobile by any means.nor is HMC while Wide-IO ,the solution targeted at mobile, is somewhat dead.
              Mobile will go with the SoC and DRAM in the same package with FO-WLP in 2017-2018. Beyond that,more advanced packaging solutions are needed.

            • Lazar Prodanovic
              September 27, 2016

              Do not take a good lords name in vain.

              Market demand does correspondence the pricing but not manufacturing costs.

              Won’t believe but currently prices are down to 1.30$ per GB & for both types on desktop naturally.

              I perfectly good know what POP is.
              You are totally wrong HBM is perfectly suitable for mobile & can bring up to 10x power savings compared to the DDR4 L JEDEC spec. A

            • realjjj
              September 27, 2016

              lol you are a clueless and insane troll. the amount of nonsense you always vomit is outstanding.

              the spot market is at 4.2$ per 1 GB DDR4 and BTW DDR4 does 2ch x16 per die.

            • Lazar Prodanovic
              September 28, 2016

              As always I miss but I miss less than you. By the way I at least stand behind my misses what can’t be said for you boy.
              http://images.anandtech.com/doci/10438/dramexchange_comparison_H1_2016.png?_ga=1.92881501.1525869238.1475022644
              http://www.extremetech.com/wp-content/uploads/2015/01/DRAMs.jpg
              http://www.extremetech.com/wp-content/uploads/2015/01/Memory-Bandwidth.jpg
              Keep in mind never the less that wide IO & HBM are mostly the same but differently configured (number of dies & wide of them [64/128 bit]) to be suitable for mobile/mainline (desktop).

              By the way fuck you!

            • realjjj
              September 28, 2016

              Because you can’t even Google and you miss the fact that DRAM prices are up.
              Are you 80 and suffering from dementia? I am actually asking because you certainly behave like that and you are only getting worse.

            • Lazar Prodanovic
              September 28, 2016

              No I am not as I didn’t forgotten any of your horseshits & there whose plenty. Do you want me to find yours trolling back at Anandteach when they did Meizu Pro 6 review? & show everyone how demented litle trol you are & un objective MTK fun boy, actually just a boy.

  4. Aeonia
    September 27, 2016

    Don’t expect the X30 anytime before next year around June, July

    • Marco Lancaster
      September 27, 2016

      It is very likely to debut in a Meizu flagship, being 7 or Pro 7. Almost no Helio X25 phone yet too…

      • balcobomber25
        September 27, 2016

        Meizu Pro 6, Xiaomi Redmi Pro, LeEco Le 2 and Elephone P9000 Edge have all used the X25.

        • Marco Lancaster
          September 28, 2016

          Yeah… I should have said few, was pretty sure that some were using, Meizu being the debut brand since always. But I would risk Elephone from the count xD, elephone is see to believe.

  5. Nolan
    September 27, 2016

    P20
    – Lower low end devices (HomTom, Bluboo etc. the 60-70 USD market)
    – Should be able to boot up the phone and keep maybe 2 or 3 apps.
    – No GPS / no GPU / No advanced connectivity

    P25
    – Low end devices masqueraded as performance editions (90-120 USD market)
    – Same as above, but will be fitted into “Plus” or “Pro models” to fool users

    X30
    – Entry level mid-range devices (120-170 USD market)
    – Mediocre mid-range performance, can handle multitasking somewhat.
    – Poor gaming, decent GPS, mediocre connectivity

    Don’t hold your breath. 2016, QualComm won with the SD.65x series.

    Come 2017, QualComm will release another mid-range chip which will be better in every way than the MediaTek top chip, while being cheaper through either Xiaomi or OnePlus or a different manufacturer. (Just like SD 650/652 vs. that crappy X20 and X25 which are also overpriced).

    • balcobomber25
      September 27, 2016

      Nolan – Qualcomm Fanboy.

  6. afghanPower
    September 27, 2016

    nm != mm
    (edit): Love your reviews dude!

    • Zi Jin Cheng
      September 27, 2016

      Edited, thanks.
      Also thanks! Hope to do many more!

  7. E8hffff
    September 27, 2016

    Mediatek is too slow to release enough SoC’s and or Chinese manufacturers want to keep making phones based on MK65xx and MK6735 series chips, so there’s no flair with this news at all. No to mention which Chinese phone maker has implementing 4K video or even slow motion since it was introduced in Mediatek SoCs. Ridiculous.

  8. realjjj
    September 27, 2016

    To add key features you miss.
    X30:
    GPU Imagination 7XTP MT4 @ 820MHz
    UFS 2.1 support , LTE Cat 10 , sensor hub, better audio, external 2×2 wifi ac
    They claim things like
    CPU 43% higher perf and 53% lower power vs X20
    100% higher memory BW and 50% lower power
    GPU 2.4 times the perf and 53% “power saving”
    Assuming lower power would be on a per same task basis.

    And P25 pushes the CPU clocks to 2.5GHz vs 2.3GHz for the P20.

    The X30 is a proper high end, we’ll see if they got an X35 at 3GHz or X35 will be 2.8GHz.
    CPU perf should be great, GPU perf very good but not class leading. Given the config and clocks, the GPU should be at some 210GFLOPS so maybe some 15-20% bellow the latest iphones- Apple uses 6 cores at lower clocks.
    No major missing parts , it could be better here and there but it’s a big step forward , closing the gap with what others have in the very high end. That’s how it goes, step by step , they can’t take the lead with the first try.

    BTW, they haven’t unveiled the SoCs, these are leaks.

  9. realjjj
    September 27, 2016
    • Yep
      September 27, 2016

      I have no idea how they could differentiate the X35 now SoC are so powerful. I hope it is true about a universal (ARM+X86) desktop UI operating system from Google that can start to make use of this power. Then 16GB RAM SoC with Quad core 3.5 GHz which are possible will make sense because they can be in convertibles and smartphones which dock into a monitor. Otherwise this is getting stupid.

    • realjjj
      September 27, 2016

      Chromebooks can run Android apps. That means that we’ll see Android apps aimed at desktop.
      So it’s normal for ChromeOS to become just a mode in Android , like Android Auto mode or VR mode.
      CPU perf is enough in phones now and can get much better. GPU perf has a long way to go.
      On 7nm in 2018, A73 could be clocked north of 3.5GHz in phones, especially if advanced packaging is used.

      A73 is a very small core, a quad cluster with 2MB L2 cache should be about 5mm2 on 10nm. I fully expect a much bigger core from ARM soon, likely targeting 7nm in 2018 and that would be a big perf boost.

      VR mode and glasses in general must be factored in too.That’s more than desktop , much more and power will be a lot more limited than in phones.

    • Guaire
      September 27, 2016

      I guess X35 will have that clock speeds and actual X30 will be underclocked. Probably ~2.3GHz for the A73 and ~600MHz for the GPU.

      They did that dirty trick with the X20/25, MT6795/6795T and almost every other chip.

    • realjjj
      September 27, 2016

      X20 was supposed to be 2.3-2.5 GHz and yes they rebranded the top SKU and added a 2.1GHz SKU.

      With dual A73 on 10nm, reaching 3GHz should be doable. Huawei seems likely to have 2.6GHz quad A73 on 16ff this year. Remains to be seen if there is an X35 and if it’s 2.8GHz or 3GHz. Given the P25 surprise, i see 3GHz as more likely. They also need to beat others in single core so they’ll push as much as they can.

      Either way, X30 wouldn’t be much lower clocks, they got no reason to do that, business wise. At lower perf the X20/25 will be half the price so even the slowest X30 SKU needs to be well above it in perf. Not that the X35 branding has much of a relevance,, it’s just a branding.

    • Guaire
      September 27, 2016

      Ares’ performance should be closer to Apple’s custom cores. Also it would fit better for dual big core designs like Mediatek’s rather than quad big clusters IMO.

      Even if stock Ares can’t match with it, a customized by ARM one might be closer if their any customer would be interested in.

    • Guaire
      September 27, 2016

      After Geekbench 4’s release, probably it won’t have a hard time at least matching Snapdragon 830’s single core performance assuming it will came with an overclocked Kryo.

      I think even a bit faster than 3GHz doable with 10nm A73.

      Strangely at both version of Geekbench, A72 doesn’t perform parallel to its higher clock speeds of both Kirin 955 and Helio X25 comparison to Snapdragon 650/652.

    • realjjj
      September 27, 2016

      Yes , Ares or Ares like as that leak is old and some things might have changed since.

      The CPU perf is enough, the gain in normal usage is small at this point as you depend more on storage and connectivity perf.
      2x big core would be great for marketing but it’s a bit pointless and 4x A7x would be more efficient with no or little perf penalty in real world usage.By 2018 an A72 followup could offer 2x the perf of an A72 at 2.5GHz and that’s quite plenty.
      But marketing matters so the best we can hope for is 2x big cores that are almost always sleeping with 4xA7x at medium clocks doing the demanding work while small cores do 80% of the work.

    • realjjj
      September 27, 2016

      Geekbench 3 total is heavily impacted by encryption and especially memory perf. Look at integer and FP scores.
      In 4 , both encryption and memory have a much lower impact but memory still distorts the total a bit. Not 100% that’s the main issue with scaling in GB 4 but has to be part of it. Looking at each individual test and doing the math would provide a good enough answer but that’s a bit of work. Could also be that in some tasks the memory bandwidth holds it back a bit at high clocks.

      I expect SD830 to use a slightly custom A73 and be branded as Kryo but with the note,” Built on ARM Cortex technology”. If they don’t have a much better Kryo, it just makes no sense to stick with it as it loses even vs A72 in perf and power.

    • Guest
      September 27, 2016

      look who just came outta their CAVE !

      *applause*

    • realjjj
      September 27, 2016

      What can i do if the new phones are boring as hell and there is nothing interesting. I’m around, just nothing worth any attention.

    • Guest
      September 27, 2016

      ahaha … just messing with you buddy

      honestly missed ya !

    • realjjj
      September 27, 2016

      Been a bit more focused on electric cars and studying Tesla lately, it’s a bit less boring than phones with 1080p A53 and same looks as in 2012.

      And the times are interesting there as oil is about to go away.

    • Zi Jin Cheng
      September 27, 2016

      I would definitely like to see more phones with Snapdragons in them, like the Zuk Z2

    • Guest
      September 27, 2016

      lel … whaddya kno !

      i’ve been gunning for an electric car for some time now !

      i WOULD LOVE TO OWN ONE IN PAKISTAN !

      any cheap ones coming to my region ?!

    • realjjj
      September 27, 2016

      The only electric car worth considering is Tesla Model 3 at this point but at 35k$ it is not cheap. Hard to figure out what China has to offer here so not factoring those in.

    • realjjj
      September 27, 2016

      Electric cars today ask you for a 10-15k$ premium over gas equivalents. Tesla doesn’t do that. By investing in what matters to enable lower costs and through product design, they can compete (and win) with gas powered cars in a certain segment. Model S is very costly but competes directly with gas powered cars in it’s segment .Last year in the US it had 25% share in the luxury segment, the green factor matters to some extent but it really is an awesome car.

      Model 3 will go head to head in the 35k $ segment, against BMW 3 Series, Audi A4 or Mercedes C-Class.

      Electric means low center of gravity, large crumple zone in front, lots of cargo space (front+back),instant acceleration, lower noise, more reliable powertrain and much cheaper fuel. So by being able to be price competitive, they actually offer better value.
      Charging at home is an advantage while for longer trips it does take longer to charge so it is a disadvantage but chargers are getting faster all the time. Today folks stay 30 mins at a Supercharger to go from 20% to 70%, tomorrow those 30 mins will be 15.
      Aside from that, Tesla also has tech enabled convenience features and the Superchargers network giving it an advantage over others. They also have the best Level 2 autonomy on the market and are well positioned to reach full autonomy soon enough and ahead of others.

      Model 3 looks great even with that weird nose. EPA range rating should be at least 215 miles, they aim for 5 stars safety and should do 0 to 60 in under 6 seconds. The glass roof offers a very different experience. No instrument cluster means that they are evolving the “UI” further and they’ll either have a HUD or a display on the steering wheel. Tesla is getting rid of the retro buttons and AC grills, they are modernizing the UI with touch and voice.

      But all that is not the most interesting part.The interesting part is car as a service with autonomous vehicles. Such a service will be much cheaper than owning a car and with some luck it all starts in 2018-2020.That’s what will make the difference and force a fast shift to electric while providing much safer and more affordable transportation.

    • Lazar Prodanovic
      September 27, 2016

      “The X30 is a proper high end, we’ll see if they got an X35 at 3GHz or X35 will be 2.8GHz.
      CPU perf should be great, GPU perf very good but not class leading.”
      Let’s disassemble those claims & see what we could really expect.
      The CPU; A73 vs A72= 10%+ performance per clock & 20% less power consumption, 10nm vs 16nm= around 25% less power usage. So per clock (A73+ 10nm)=110% performance and 60% power consumption. With a clock difference 2.3<2.8=22%, 110×122%=135% performance & up to 10% power saving (I am generous hire). Mid cluster will be spending the same amount of power & deliver 15% performance bump, small cluster won't save anything (52% power per clock, 33% higher clock) but it will be marginally faster (13%).
      RAM: doubling the frequency & going to DDR 4 (2xDDR 3 32 bit vs 4x DDR4 16 bit) won't really duble true output, additional latencies will hurt CPU performance & output will help GPU, estimated bandwidth gain is in the best bet margines of 60%.
      GPU; PowerVR Rogue 7XTP MT4 vs MALI T880 MP4 per clock around 40% more performance & equally more power consumption. 17% clock boost & the given 10nm advances brings us to the 42% bigger power consumption & adding a up to 20% performance gain from 60% bigger memory output brings us to (117%x120%=140.4) 140.4×1.4=196.56%. Not bad considering GPU performance but not to good considering power consumption either. Anyway this brings GPU to a category of being capable to support 2K resolution displays or capable for VR with FHD display resolution. Either way I am not happy, not that it would be much better if they did go With G71 aka T980 MP6 considering cost/performance/power consumption metrics (25% bigger & 30% faster wile consuming 20% more power [but if down clocked 30% so that performance is the same consumption would be 50% or compared to the 7XTP MT4 only 60% & comparable to the previous gen X20 I would call this a win, win solution]). If we include royalties & licensing costs probably the size difference of 25% cost for T980 MP6 (G71) would come to some 10% cost difference to manufacture & with ARM POP G71 libs to 0%.
      So what is to expect?
      Well considering X20 couldn't keep up max frequencies in demanding CPU or GPU tasks for a longer period of time & how the X30 doesn't bring any significant power savings compared to it just on CPU front things will mostly remain the same, on the other end GPU will use 42% more power so it will be much worser there. We also sow how core pilot is working bad in the X20 deca core so it will continue to be a sama story with X30. All in all a really bad design from my view point. Thing that pissed me the most is that they still used dual M4 CPU cluster for sensors hub processing (so this is basically the 12 CPU quad cluster SoC) instead employing the same ISA V8 64 bit A35's as then they could be used as general purposes (all tasks) ones. Considering that the X30 SoC will cost around at least 35% more & that it will even more rise the final product cost as it will must employ 4x DDR 4 chips (instead of two DDR 3 ones 4GB) that will cost 2.5x (4GB) to 4x (8GB) times more I seriously doubt it will be commercially highly successful product.
      I am wondering how complicated it would be to make a turbo boost reserved clock frequencies on a ARM reference big cores like A73? As this would eliminate need for much things. Probably not to hard as Samsung is doing it (but not so good).

    • Guest
      September 27, 2016

      nice description !

      im definitely interested in autonomous vehicles … but i just wanted a cheap car to drive to uni (1.5 hours by bus and less than 30 min if i drive by car)

    • Lazar Prodanovic
      September 27, 2016

      Well their is an need for the developer & ultra book class SoC & I think reference Quad core A73 on 10nm @ 2.8GHz consuming 1.200mW per core or under 4.5W in full usage on POP ARM lib just qualifies to brick Intel’s M offering considering both metrics performance & performance per W (as 2.8GHz A73 quad core is equivalent to the top M sku at 1.6GHz). Remains to implement the boost & turbo frequency implementations on ARM & to do it properly (in hardware).

    • realjjj
      September 27, 2016

      Not really but will only address the DRAM cost claim because it’s by far the most ridiculous. DRAM sells on a per GB basis and you can fit a 2x2x2 config in a PoP, like Samsung does here https://www.chipworks.com/about-chipworks/overview/blog/samsung%E2%80%99s-galaxy-note-7-is-more-the-batteries

    • realjjj
      September 27, 2016

      Other electric cars offer shitty value that’s why they are not worth it at this time.
      Niu has some new scooters, not sure if you’ve spotted those yet.

      Why don’t you get a pedelec https://en.wikipedia.org/wiki/Pedelec
      It’s a bike with an electric motor that boosts your pedaling. So you work a bit to stay in shape but a lot less effort and you got nice range. If you run out of juice ,no problem.
      Xiaomi’s pedelec is not great nor all that cheap but it’s on option i guess.

    • Lazar Prodanovic
      September 27, 2016

      Mediatek didn’t done POP of the X30 for memory nor they need to & that would actually only cost them more & make costumers unhappy (as they wouldn’t have options to pick RAM by them self [including amounts of it]). Samsung is only doing it for SSD’s, the M2 specification for the bord is small so they gain for nand dies that way. It remains to be seen on commercial SoC’s & it only makes sense for HBM2 (or newer) integration.
      I didn’t go with a claim that the 10nm manufacturing process is much pricier per transistor but did go with a claim that 40% bigger GPU counts in for around 22~25% more transistors on the whole SoC & certainly yields won’t be fantastic on the new & un perfected 10nm FF+ TSMC node so my assumption of 35% more cost is actually optimistic one.
      Samsungs 10nm FinFET node is much more mature (&still far from good yields) but also considerably worser than TSMC.
      Main problem with X30 RAM controler is that it is 4×16 bit so you see you have to use 4×1 chip’s at least (to utilize quad controller) & that costs more than going only with two (as on X20).

    • realjjj
      September 27, 2016

      Seems that you don’t know what PoP is. PoP means package on package, it’s been the default for a long time.

      The config you see in the note 7 is 4Gb x 8 for 4GB RAM, nothing special on the cost side and DDR4 prices are quite ok That’s why even the P20 went DDR4.

    • Lazar Prodanovic
      September 27, 2016

      It’s a pre rooted lib of a functional block for a given lithography process. & if you implement it in the same chip (SoC) then it’s understud that you will have to go with the same manufacturing process, it also means that your SoC just become a Mammoth sized class one (considering how much RAM takes in transistor count especially if we speak about large amounts of it), it also means that the chances for wafer production mistakes are bigger by an order of magnitude. It costs more A: you will have more un functional dies, B: you can buy a RAM chips on older manufacturing processes cheaper.

      Igi Pop.

    • Marco Lancaster
      September 27, 2016

      It’s definitely boring lately..

    • realjjj
      September 27, 2016

      lol we are talking PoP not ARM’s POP IP
      It’s just a pretty boring package for DRAM at a very normal cost.

      https://en.wikipedia.org/wiki/Package_on_package

    • Lazar Prodanovic
      September 27, 2016

      & MTK doesn’t produce RAM.
      So get back to the reality.
      In reality it costs more & makes a sense only with HBM (for now).

    • realjjj
      September 27, 2016

      You really don’t understand this conversation at all.

      You made a claim that is beyond absurd:
      “the final product cost as it will must employ 4x DDR 4 chips (instead of two DDR 3 ones 4GB) that will cost 2.5x (4GB) to 4x (8GB) times more Iseriously doubt it will be commercially highly successful product.”

      And now you are utterly lost. The DRAM is in a package using conventional wire-bond.You got a bunch of dies, a normal substrate and solder balls and, with PoP, you just stack the DRAM package over the SoC and that has been the default for a long time.

      DRAM sells on a per GB basis and, outside of extreme capacities on a per die basis , the size of the die doesn’t make much of a difference. DDR4 prices are very good compared to DDR3 , not a problem at all, even less so in 6months or more from now when X30 ships.If anything DDR3 would cost more.

      As for how to deploy LPDDR4 , maybe this helps you http://semiengineering.com/using-multi-channel-connections-for-optimized-lpddr4-power-performance/

    • Lazar Prodanovic
      September 27, 2016

      Ok it will still remain pricier than DDR3 & you will still need to use 4 packages at least for 4 controllers. It may not cost duble but it will cost more. DDR3 can’t cost more as its easier to manufacture & it still isn’t the obsolete product nor it will become that in the long time. Even DDR3 is marginally not usable for anything that is latency sensitive (not to mention real-time) like ray tracing, SQL search & cetera… DDR4 is much worse naturally so their is a real need for a switch to something else (like HBM). Again HBM is much bigger candidate for a PoP as the memory controller is on it (& will stay there)…

      So my claim is just exaggerated wile you are absurd (in many ways as PoP & MTK).

    • realjjj
      September 28, 2016

      OMFG
      It’s a single package- apparently you don’t even know what a package is…

      It’s safe to assume that it is a standard 366 ball memory package, Maybe look at the freaking LPDDR4 standard on wiki if nothing else.
      Pricing for memory depends on supply and demand balance not on costs. DDR4 in desktop today has a less than 10% premium over DDR3 and folks are still ramping 20nm DDR4 but very soon DDR4 becomes cheaper. Today 1GB LPDDR4 is likely at about 4.5-5$.
      In mobile the DDR4 bandwidth is needed and DDR4 offers lower power per bandwidth and that’s critical when every mW matters.

      You still have no clue at all what PoP is. Anyway, HBM is not a solution fit for mobile by any means.nor is HMC while Wide-IO ,the solution targeted at mobile, is somewhat dead.
      Mobile will go with the SoC and DRAM in the same package with FO-WLP in 2017-2018. Beyond that,more advanced packaging solutions are needed.

    • Lazar Prodanovic
      September 28, 2016

      Do not take a good lords name in vain.

      Market demand does correspondence the pricing but not manufacturing costs. It’s logical to expect a rise of the demand for DDR4 & a prices with it.

      Won’t believe but currently prices are down to 1.30$ per GB & for both types on desktop naturally.

      I perfectly good know what POP is.
      You are totally wrong HBM is perfectly suitable for mobile & can bring up to 10x power savings compared to the DDR4 L JEDEC spec. HMC is not suitable for mobile nor general purposes. But it’s OK to be confused as the amateur that you are.

      & when it goes it will be with a PoP HBM package & probably from Samsung. Certainly not from someone who doesn’t manufacture memory like MTK.

    • realjjj
      September 28, 2016

      lol you are a clueless and insane troll. the amount of nonsense you always vomit is outstanding.

      the spot market is at 4.2$ per 1 GB DDR4 and BTW DDR4 does 2ch x16 per die.

    • balcobomber25
      September 28, 2016

      Current estimates say there around 53 years of oil left on the planet. Some analysts say that number is closer to 100 as new oil fields and drilling tech (like fracking) is being explored.

      Electric cars are very interesting but they are decades away from being mainstream.

    • Lazar Prodanovic
      September 28, 2016

      As always I miss but I miss less than you. By the way I at least stand behind my misses what can’t be said for you boy.

    • Lazar Prodanovic
      September 28, 2016
    • Lazar Prodanovic
      September 28, 2016

      Keep in mind never the less that wide IO & HBM are mostly the same but differently configured (number of dies & wide of them [64/128 bit]) to be suitable for mobile/mainline (desktop).

      By the way fuck you!

    • realjjj
      September 28, 2016

      Because you can’t even Google and you miss the fact that DRAM prices are up.
      Are you 80 and suffering from dementia? I am actually asking because you certainly behave like that and you are only getting worse.

    • realjjj
      September 28, 2016

      That’s the opinion of clueless people with zero vision
      Tesla already offers better value than gas competitors today but that’s not the point as it is not about car sales anymore.
      As i mention bellow, car as a service with autonomous vehicles will force a very fast transition to electric. CaaS is much cheaper and more convenient.

      The day regulators ok full autonomy , oil hits a wall and there will be a catastrophic decline in demand over just a few years not decades. It all starts in 2018-2020. By 2025 even folks in developing nations could afford to pay 50$ per month to ride in a rather nice car.
      Oil is done.

    • Lazar Prodanovic
      September 28, 2016

      No I am not as I didn’t forgotten any of your horseshits & there whose plenty. Do you want me to find yours trolling back at Anandteach when they did Meizu Pro 6 review? & show everyone how demented litle trol you are & un objective MTK fun boy, actually just a boy.

    • Guest
      September 29, 2016

      wow … wait.
      electric motor and pedals.

      won’t i just kill my feet if i pedal from home to uni and back ?!

      is this good for long distances …
      its over 13 Kilometers !!!

      im was looking at scooters but parents are not so open about the idea since the Traffic in Karachi is a bit dangerous for bike riders.

    • balcobomber25
      September 29, 2016

      That;s the opinion of science and business. Unfortunately, in your little $50 a month dream, it is not practical or based on anything other than your dreams. Auto Industry experts estimate that by 2040 electric cars will have 35% market share. The problem is it will take another decade at least to convert existing engine manufacturing facilities. It will take a decade for companies to convert their current models. On top of all that is the costs associated with such a move, it won’t be cheap.
      It will cost Auto companies billions to completely stop selling gas powered cars. And then there is Big Oil who provides auto companies with funding. Oil isn’t going anywhere anytime soon, auto companies will use it till it runs dry. Neither of us will see a point in our lifetime when there are no gas powered cars on the road.

      While your opinion is a good one, it is not a practical one.

  10. Boni M.
    September 27, 2016

    Don’t expect the X30 anytime before next year around June, July

    • Marco Lancaster
      September 27, 2016

      It is very likely to debut in a Meizu flagship, being 7 or Pro 7. Almost no Helio X25 phone yet too…

    • balcobomber25
      September 28, 2016

      Meizu Pro 6, Xiaomi Redmi Pro, LeEco Le 2 and Elephone P9000 Edge have all used the X25.

    • Marco Lancaster
      September 28, 2016

      Yeah… I should have said few, was pretty sure that some were using, Meizu being the debut brand since always. But I would risk Elephone from the count xD, elephone is see to believe.

  11. Guest
    September 27, 2016

    P20
    – Lower low end devices (HomTom, Bluboo etc. the 60-70 USD market)
    – Should be able to boot up the phone and keep maybe 2 or 3 apps.
    – No GPS / no GPU / No advanced connectivity

    P25
    – Low end devices masqueraded as performance editions (90-120 USD market)
    – Same as above, but will be fitted into “Plus” or “Pro models” to fool users

    X30
    – Entry level mid-range devices (120-170 USD market)
    – Mediocre mid-range performance, can handle multitasking somewhat.
    – Poor gaming, decent GPS, mediocre connectivity

    Don’t hold your breath. 2016, QualComm won with the SD.65x series.

    Come 2017, QualComm will release another mid-range chip which will be better in every way than the MediaTek top chip, while being cheaper through either Xiaomi or OnePlus or a different manufacturer. (Just like SD 650/652 vs. that crappy X20 and X25 which are also overpriced).

    • balcobomber25
      September 28, 2016

      Nolan – Qualcomm Fanboy.

  12. Tommi
    September 28, 2016

    X35 most likely released also next year, higher clocks on CPU + GPU.

  13. Zdeněk Anděl
    September 28, 2016

    I hope x30/35 GPU will be better than x20/25 bcuz GPU is what those SoCs kills

    • highwind
      September 29, 2016

      Because everybody plays Smartphone games all day long!
      Oh wait…

      • Zdeněk Anděl
        October 2, 2016

        I do actualy 😀
        Don’t get me wrong x20 on LeEco le 2 is great…just not for games 🙂

  14. Guest
    September 28, 2016

    X35 most likely released also next year, higher clocks on CPU + GPU.

  15. Zdeněk Anděl
    September 28, 2016

    I hope x30/35 GPU will be better than x20/25 bcuz GPU is what those SoCs kills

    • provokanter Tabellenführer
      September 29, 2016

      Because everybody plays Smartphone games all day long!
      Oh wait…

    • Zdeněk Anděl
      October 2, 2016

      I do actualy 😀
      Don’t get me wrong x20 on LeEco le 2 is great…just not for games 🙂