Today, Qualcomm announced a number of details in the Snapdragon 820, specifically about their Kryo CPU. Given that the Snapdragon 810 was a somewhat standard 4x Cortex-A57/4x Cortex-A53, it was clear that that this chip was a stop-gap for a future fully custom design. With the Snapdragon 820 announcement, the first major bit of information that we received was that this would be a return to a custom CPU core design, and today Qualcomm is finally unveiling a bit more information on Kryo.

The two main spec details that are being disclosed today is that the quad-core Kryo CPU in Snapdragon 820 will reach up to 2.2 GHz, and that the SoC will be manufactured on Samsung’s 14nm FinFET process. It isn’t stated whether this is the 14LPP process, which will give up to 10% transistor performance improvement over 14LPE which was seen in chips like the Exynos 7420, but it’s a safe bet that it is. As a result of the new architecture and new process node, Qualcomm is claiming up to a 2x increase in performance and up to a 2x increase in power efficiency compared to Snapdragon 810.

The final part of this announcement is Symphony System Manager, which is said to be designed to deal with heterogeneous compute in an efficient manner. This is likely to be a kernel-level mechanism that ensures that the SoC is well-optimized for use in a smartphone or any other application. Given the focus on heterogeneous compute for this launch, I wonder if Qualcomm is going for some form of heterogeneous CPU design as well.

Source: Qualcomm

POST A COMMENT

60 Comments

View All Comments

  • zodiacsoulmate - Wednesday, September 02, 2015 - link

    2x say what Reply
  • syxbit - Wednesday, September 02, 2015 - link

    The SD810 was TERRIBLE, so 2x SD810 should be doable. Especially if their internal benchmarks stress test (since SD810 perf drops significantly under heat) Reply
  • tuxRoller - Wednesday, September 02, 2015 - link

    It wasn't THAT bad. If it doubles performance (in what area?) in geekbench that would give it a single core score of around 2600-2800(or WELL into high powered Intel territory). To be clear, I'm not expecting that. What I do expect is it scoring around 1900 for a single core, and 6000-6500 (closer to the lower end of that range) for all four Reply
  • jjj - Wednesday, September 02, 2015 - link

    Up to 2x when you factor in the work on DSP and GPU. But what exactly will use GPGPU is unclear ,especially if the DSP does image manipulation.
    When a company has a good product there is no need for vague and misleading numbers so i see this as a red flag. Hoped for 60-80% gain over a cool SD810 but some rumors of 35% CPU gain and this today make me wonder if Kryo is slower than what A72 is supposed to deliver, if A72 can reach it's targets. If Kryo and A72 are close, power and die area can make a difference so we'll see. Messing this one up would be a multi-year problem and given the lack of competition in the high end it wouldn't be all that great for us.
    Reply
  • ZolaIII - Thursday, September 03, 2015 - link

    I think they ment general purpose CPU cores performance & I am certain that 2x gain is overstatement same as we can say that cortex A53 is up to 2x faster in some tasks than A7 but overlay performance gain is about 35%. Spectra which is basically just a large SIMD aria (2x1024 bit) (a bit similar to AVX in purpose at least) do most more frequent FPU - VFP tasks helped bi DSP in those more paralel, basically most multimedia stuff. Knowing how Qualcomm mixes smaller core logic with bigger ones cutting bigger ones I am not convinced it will be faster then A72s but it will probably be more power efficient. We will see soon enough. Reply
  • jjj - Friday, September 04, 2015 - link

    The quote is "With Kryo CPU and Snapdragon 820, you can expect up to 2 times the performance and up to 2 times the power efficiency when compared with the Snapdragon 810 processor."
    That's not a phrasing you use if you mean CPU and there is a strong emphasis on the DSP and in their announcement.
    Looking at it another way, why push expectations that high and then fail to deliver (seems pretty hard to actually deliver on 100% CPU gain). If you have a great product you still want to surprise on the upside at launch so this just feels off.
    Reply
  • edzieba - Thursday, September 03, 2015 - link

    UP TO 2x. which means when you draw the performance/power curves for both, there is at least one point where you can say "at this power, X performs at twice Y" and one point where you can say "at this performance level, X is using half the power of Y". Reply
  • michael2k - Thursday, September 03, 2015 - link

    Sure. Have you seen the HTC M9? http://www.anandtech.com/show/9102/the-htc-one-m9-...

    The 810 wasn't actually all that great under benchmarks. Doubling it's performance means it is still likely to be less powerful than next week's iPhone 6S.
    Reply
  • lilmoe - Wednesday, September 02, 2015 - link

    "I wonder if Qualcomm is going for some form of heterogeneous CPU design as well"

    That would be pretty interesting actually.
    Reply
  • TheJian - Wednesday, September 02, 2015 - link

    The problem is WHEN. It seems Nvidia is first this xmas (for nov devices I'd say, with apple sept 9th), as Qcom seems to be quite late with this. Note there is no date on this pic for when it's expected. With a 14nm Finfet chip, I'm wondering if NV can use their own modem in some cases, and like Intel use Qcom for MU-Mimo if desired (see recent chromebooks with Intel pairing qcom). With power dropping so much from die shrinks, there isn't much need for on-die crap now if NV's old 150 modem can't get the job done. I'm wondering if Nv could get into top end phones even if they have to use Qcom modem (or samsung, considering the suit, maybe they'll get a deal).

    Either way, Qcom is late, which is why samsung dumped them and why their stock has plummeted over the last year (along with china cheapo socs/currency manipulation etc hurting stock price too).
    Reply

Log in

Don't have an account? Sign up now