Comments Locked

36 Comments

Back to Article

  • SydneyBlue120d - Wednesday, November 14, 2018 - link

    Can we finally have unlimited 2160p60 encoding?
  • PeachNCream - Wednesday, November 14, 2018 - link

    No.
  • darkich - Wednesday, November 14, 2018 - link

    Andrei, it looks to me like you're making a surprisingly illogical and ironical mistake in your expectations based on Samsung's CPU improvement claims.

    Ironical because you always stress that performance is about real life and sustained metrics and not the absolute claims.
    Samsung's claims are most certainly referring to peak theoretic performance (that should, btw, absolutely be superior to Kirin 980), but the key thing is, this time the real life/sustained performance should be *dramatically* improved compared to the predecessor.
  • Andrei Frumusanu - Wednesday, November 14, 2018 - link

    > Samsung's claims are most certainly referring to peak theoretic performance (that should, btw, absolutely be superior to Kirin 980

    And why would you be so certain of this? The Kirin 980 is 8-15% faster than the 9810 while using half the energy. I'll have the review on this up soon enough. Samsung's 20% improvement *or* 40% efficiency gain claims are simply not enough.

    If the absolute claims don't pan out, there is little chance the real life claims will.
  • Gondalf - Wednesday, November 14, 2018 - link

    Half the power? So faster and so frugal? 5nm??
    Umm there are voices that very recent phones, iPhone XS included, are a delusion in real usage power consumption, worse than older generation phones.
    Why to go on 7nm if your phone give you bye bye in a short time. For few more fps on a game??
    Likely better wait new Galaxy with all the power management features on to judge.
  • MrSpadge - Wednesday, November 14, 2018 - link

    > Umm there are voices

    There are voices for anything on the net.
  • darkich - Wednesday, November 14, 2018 - link

    ..why am I certain??
    Um, because that's how these claims always work.
    Silly question.

    Let me rephrase..the Kirin 980 is faster in your benchmarks where Samsung's 9810 was throttled down to basically half of what it theoretically should've been capable of.
    Samsung basically only needs to make sure their chip actually works normally this time, nevermind the theoretic peak value improvement numbers.
    And if that turns out to be the case (given the *major* conceptual changes done by Samsung this time around, I'd say it is realistic to expect the chip to be far more balanced at least) Kirin should simply be easily outperformed.
  • Andrei Frumusanu - Wednesday, November 14, 2018 - link

    > where Samsung's 9810 was throttled down to basically half of what it theoretically should've been capable of.

    Uhm no it's not throttled. I measure at max frequency. At this point I'm not sure you're even differentiating between actual CPU performance and the real world dynamic workloads.
  • darkich - Wednesday, November 14, 2018 - link

    Alright then, my bad.
    But care to explain why the 9810 outperforms kirin in geekbench and not in most of other benchmarks?
  • LarsBolender - Wednesday, November 21, 2018 - link

    Read the articles, he already explained it within those.
  • levizx - Wednesday, November 14, 2018 - link

    Wow, just wow. The first thing you should know is, single threaded benchmarks are never throttled, unless it's running for 5+ minutes. Certainly never ever seen a non-endurance benchmark where the CPU is throttle to 50% peak performance.
  • darkich - Wednesday, November 14, 2018 - link

    I'm confused.. my point was kinda exactly that..exynos beats kirin in geekbench single, and comes roughly at par in multi core, but why not in other benchmarks?
    Because it's not allowed to run at full power(!?)
    With the new setup, I don't see why they wouldn't be better across the board.
  • Wilco1 - Wednesday, November 14, 2018 - link

    They are close: https://browser.geekbench.com/v4/cpu/compare/10394...
    Exynos-M3 is better at throughput benchmarks, especially FP, and its lower memory latency must help there too. On the other hand Cortex-A76 wins all branchy benchmarks by a good margin (eg. LZMA, LLVM, Dijkstra), and that is likely why it does well on SPECINT.
  • Dmcq - Wednesday, November 14, 2018 - link

    "Arm doesn’t allow third-party cores to plug into the DSU." Why on earth not?
  • levizx - Wednesday, November 14, 2018 - link

    Because.
  • Teckk - Wednesday, November 14, 2018 - link

    This will go against the next Snapdragon flagship? Also, how does 8 LPP compare to Intel 14nm?
  • saratoga4 - Wednesday, November 14, 2018 - link

    Isn't 8LPP Samsung's version of TSMCs (nonEUV) 7nm node? Surprised they don't see more improvements.
  • s.yu - Wednesday, November 14, 2018 - link

    No, it's seen as a stopgap because their 7nm didn't make it in time, it's actually pretty close to their 10nm.
  • eastcoast_pete - Wednesday, November 14, 2018 - link

    @Andrei: Thanks! I look forward to the Mate 20/Kirin 980 review with the performance and performance/Wh numbers you mentioned.
    Question: Any word from Samsung if they can run the big ("M4") and the not-as-big (A75) cores simultaneously (heterogeneous multicore, would be quite a feat in itself), or will it run each group of cores exclusively of each other?
    Lastly, if the "M4" is indeed inferior to the A76 in performance and perf/Wh, did Samsung commit the sin of throwing good money after bad? In other words, is their license to customize ARM cores plus development costs for their custom CPU core really that much cheaper than an outright license to just plain use A76 cores? They obviously are using ARM stock designs for the A55, A75, and the Mali cores, so the not-invented-here syndrome cannot be the whole answer.

    Unrelated to this, I look forward to Qualcomm's next flagship design. Their newest mid-level chip uses two A76 cores, and these are, in the moment, probably the most powerful cores in any Qualcomm chip, including the 845. So, QC has some serious work on their hand, unless they simply go for quad-core A76 for their big cores and their customization is mainly the "Qualcomm" logo on their new chips.
  • Javert89 - Wednesday, November 14, 2018 - link

    Cortex A76 are used as middle core on the Kirin.. We wait for your review but along the lines of Huawei it seemed that at 1.92 ghz and 0.6 watt an A76 is capable to offer the same performance of a 2.8 ghz 1.2 watt A75.. So may Samsung have got wrong choice of middle core, or an A76 as middle core on 8nm is still too big?
  • s.yu - Wednesday, November 14, 2018 - link

    I agree, no idea why they didn't choose A76, Samsung's not one to shy away from bigger dies, unlike Huawei who's never missed any opportunity to cut corners.
  • bubblyboo - Wednesday, November 14, 2018 - link

    Huawei only cuts corners on the GPU though?
  • Eris_Floralia - Wednesday, November 14, 2018 - link

    NANDs.
  • id4andrei - Wednesday, November 14, 2018 - link

    Samsung has a duty to send the first S10 off the assembly line straight to Andrei Frumusanu if it is to redeem itself.
  • The_Assimilator - Wednesday, November 14, 2018 - link

    Andrei's already made it very clear that this "new" SoC is starting behind its competitors. There won't be much redemption unless Samsung is basically willing to give 9820 away for free.
  • s.yu - Wednesday, November 14, 2018 - link

    This A75 and 8nm is really looking bad for Samsung, unless their scheduler is miraculously far superior to Huawei's.
  • lucam - Wednesday, November 14, 2018 - link

    Mali GPU is just a garbage GPU, don’t know why they don’t switch to PowerVR..
  • CityBlue - Wednesday, November 14, 2018 - link

    What are the odds on this Exynos suffering from a very significant design and/or implementation defect like many of the previous Exynos products? I've often wondered why Samsung aren't selling the Exynos to third parties (because, on paper at least, they look like interesting and competitive products) but given the number of defects and flaws it's pretty obvious why the only real user of Exynos SoCs is Samsung, as nobody else wants to take the risk.
  • s.yu - Saturday, November 17, 2018 - link

    They are, actually, at least Meizu is a regular customer of Exynos SoCs.
  • Ej24 - Wednesday, November 14, 2018 - link

    If only this would be used in the US version of the Samsung devices. No reason it can't be. My S6 on VZW has an Exynos soc. But it seems all we get here is Qualcomm snapdragon.
  • CityBlue - Thursday, November 15, 2018 - link

    Samsung probably wants to avoid a class action lawsuit in US courts when it is discovered that various features of the Exynos SOC don't work as expected (or work at all) due to design flaws, crippling performance... I mean, who the hell manages to release a SOC without a fully functioning CCI? Yep, that would be Samsung. Fire the QA/simulation department...
  • anonym - Monday, November 19, 2018 - link

    Maybe there is some discount for cdma2000 patent fee if snapdragon is on.
  • jcompagner - Thursday, November 15, 2018 - link

    i really don't get why did go for A75
    is the A76 so much bigger in size?

    Seems to me that all the high end chips of next year will be on A76, maybe the M4 can beat those, but still why then add A75 as a middle cluster?
  • Raqia - Thursday, November 15, 2018 - link

    There are interesting reports of Qualcomm using an even more aggressive configuration than either the Kirin 980 or the Exynos 9820. They are using a 1+3x(A76)+4x(A55) core configuration clocked at 2.84+2.4+1.78 ghz respectively:

    https://www.weibo.com/1778375693/H2L5sx9Rl?from=pa...

    They're onto the marketing benefits of a high single-core geekbench score.
  • tuxRoller - Thursday, November 15, 2018 - link

    First, the very latest news (11-15) suggests they may keep the 2 island approach.
    Second, is it marketing? While lots of apps are highly threaded, you still tend to see one main thread with helpers (whose loads vary immensely), and stalls on the main thread lead to jank.
  • zamroni - Thursday, November 15, 2018 - link

    It seems Galaxy S10 is not a good buy. My phone contract ending is also 2020, so better wait until S11. Maybe at that time the medium cores will be A76, instead of A75.

Log in

Don't have an account? Sign up now