POST A COMMENT

98 Comments

Back to Article

  • CiccioB - Monday, April 07, 2014 - link

    I can't understand how difficult is to make a "quick shift" of the roadmap when you are using ready on-the-shelf IP like Cortex A-57/53 while the rest of the SoC is the same as for previous projects you have thrown away. Reply
  • aruisdante - Monday, April 07, 2014 - link

    If you think that changing the entire product roadmap for two years for an enormous, international multi-billion dollar company is 'simple', then you probably don't have much experience with big-business. Reply
  • CiccioB - Monday, April 07, 2014 - link

    Actually this "quick shift" is going to require about 1.5 year on paper.
    Qualcomm is going to use ARM IP for thier SoC cores, so past wrong choices that understimated 64 bit marketing power is an economic problem in those 32bit 28nm investments that will never see the light in, not really a time related one, as 64 bit 20nm ARM IP is aready available.

    The shocking impact is to see Qualcomm not ready for the new 64 bit step after having dominated the 32bit market for some years now. And it is not only Apple (which races in its own market) their problem: nvidia and Samsung already announced their 64 bit architetures while Mediatek and other chinese/tawainese producers already have 64 bit SoC ready.

    I already forsee that Qualcomm could not continue developing an in house custom architecture for main stream market when ARM architecture would have become too complex. I think ARM v8 64 bit architecture mark the sign. Qualcomm will probably create a custom 64 bit architecture for SoC destined to other (server?) market but it will take time (quite a lot) and competitors will not wait for them. 64 bit (marketing speaking) is now, and everyone is trying to catch the passing train.
    Reply
  • mrdude - Monday, April 07, 2014 - link

    One of the biggest benefits the ARM ecosystem has is exactly that point: being able to react and adapt quickly by utilizing 'stock' designs with a significantly lower time-to-market. This model extends to the fab side as well.

    Qualcomm likely (and perhaps correctly) didn't see a need to push for 64-bit designs so early. Apple caught them off guard and Qualcomm had to react quickly. If this were to hypothetically happen on x86, we'd be waiting another 3-4 years to see the big rival push out a competing silicon.

    Hopefully Qualcomm has been working quietly on an ARMv8-a design. I'm eager to see what they can come up with.
    Reply
  • blanarahul - Monday, April 07, 2014 - link

    First 20 NM chips?? Reply
  • mikegonzalez2k - Tuesday, May 20, 2014 - link

    The purpose of a roadmap is the development of the tasks needed for the critical path. Changing those tasks completely modifies the critical path because other tasks may be dependent on the tasks which were removed from that path.

    Here is a basic example of the Critical Path Method

    https://www.youtube.com/watch?v=DdDzybQ_9vM
    Reply
  • SydneyBlue120d - Monday, April 07, 2014 - link

    This means the rumors of a Nexus 6 based on LG G3, running the 805 may be true... I really hoped to have HEVC on board acquisition for Nexus 6, but this will not be the case... Can You make a comparison with Intel Tablet/Smartphon SOC 64 bit proposition ? Thanks a lot :) Reply
  • antef - Monday, April 07, 2014 - link

    Not sure why people keep saying "Nexus 6". They name them by screen size, and I highly their next phone will have a 6" screen. Reply
  • antef - Monday, April 07, 2014 - link

    should be "highly doubt" Reply
  • Aenean144 - Monday, April 07, 2014 - link

    Just to be pedantic, Google names their phone whatever the hell they want. There's no correlation there. Prior to the Nexus 5, was the Nexus 4, and I'm pretty sure it's stretching it to say they called it "Nexus 4" because of its screen size. It's convenient to think of the number as the screen size, until it isn't.

    And you already know why people are calling it Nexus 6. It's just a convenient way to indicate they are talking about the next Nexus phone. Not sure why people just don't say the 2014 Nexus though.
    Reply
  • niva - Monday, April 07, 2014 - link

    Exactly, if they call it Nexus 6 then what will the next model be called? There's already a Nexus 7 out today.

    They need to rethink the naming of their phone. HTC already ran into it with the HTC One, perhaps sticking to a year of release is the wise choice for each model.
    Reply
  • phoenix_rizzen - Monday, April 07, 2014 - link

    You can't call it "2014 Nexus" because there are multiple Nexus devices (phone, tablet) in multiple screen sizes.

    You could call it "the 2014 Nexus phone", but that's a mouthful and a half.

    Google really screwed the pooch calling the tablets "Nexus 7" and "Nexus 10", as that just makes things even murkier. What happens when the 7th generation Nexus phone is released? Bad enough they have two "Nexus 7" devices as it is.

    Naming for Nexus devices is a mess right now.
    Reply
  • SunLord - Monday, April 07, 2014 - link

    Actually it will likely just be called the 2014 Nexus 5 just like with the newer 2013 Nexus 7 and the original 2012 Nexus 7. Google is jsut going to keep reusing the Nexus 5 Nexus 7 and Nexus 10 branding and just version it by year until they change from Nexus to something else. Hell they could just call it the 2014 Nexus and drop the number all together. Reply
  • MattCoz - Thursday, April 10, 2014 - link

    It's not stretching it, that IS why they called it the Nexus 4. Just like the Nexus 7 is 7" and the Nexus 10 is 10" and the Nexus 5 is ~5". Reply
  • phoenix_rizzen - Monday, April 14, 2014 - link

    Except that the Nexus 4 was 4.7". And the Nexus 5 was 5". Considering both of those round up to 5", the screen size was not part of the name for either of them. Also, considering the Nexus 5 is actually physically smaller than the Nexus 4 ... doesn't seem logical that the screen size would be part of the name.

    The Nexus 4 *was* the forth Nexus phone though. And the Nexus 5 *was* the fifth Nexus phone. So, unless they change the naming scheme again (which isn't all that unlikely if you look at past names), it would be logical for the next Nexus phone to be called "Nexus 6", regardless of the screen size.

    Nexus One
    Nexus S
    Galaxy Nexus
    Nexus 4
    Nexus 5

    Either Nexus S and Galaxy Nexus are aberrations in the numbering scheme, or there is no "standard" for naming Nexus phones, and Google can name the next one anything they want.
    Reply
  • jjj - Monday, April 07, 2014 - link

    The big story is how badly they messed up that they have to use standard cores.And acting fast by using standard cores is not much of a feat. It's also rather annoying that they'll have 20nm SoCs only in a year from now, guess we all kinda hoped for better than that.
    It's also rather odd that they announced it so early and well before sampling.Would be really interesting to have more details on the competition's roadmap for next year.
    Reply
  • arnavvdesai - Monday, April 07, 2014 - link

    I think the big reason Qualcomm able to move this quickly is that ARM already had a 64bit stack ready & a reference design. It is obvious that Qualcomm had nothing because it is relying on ARM for it's initial design.
    However, I must commend them on swallowing pride & going with ARM as it allowed them to get to market faster rather than wait for their designs to complete.
    Reply
  • Wilco1 - Monday, April 07, 2014 - link

    I don't think it is a matter of pride. QC has been using many ARM cores over the years - just not for their high-end. I also suspect that beating A57 performance is a difficult goal to achieve, so it will take a lot of time (A57 has about the same IPC as Apple A7 but runs at twice the frequency). Reply
  • CiccioB - Monday, April 07, 2014 - link

    Apple A7 cores are also massively big, a thing that a 3rd arty SoC producer cannot afford if not for very high end markets where the SoC cost has a meaning.

    That's why I previously said that QC custom architecture will be available in some time from now and only for a specific market. On low end device you can't compete with Mediatek or anyone using ARM ready to use architecture.
    Reply
  • dylan522p - Monday, April 07, 2014 - link

    That is complete bullshit. A57 has 20-30% IPC than A15. It doesn't rum at twice the clocks either. Reply
  • Wilco1 - Monday, April 07, 2014 - link

    Read the article, Anand says 25-55% IPC gain. I'd expect further gains from 64-bit as well (A7 showed good gains there, and those apply to A57 as well). A57 will start at ~2GHz, but quickly go to 2.5GHz (twice that of an iPhone) and beyond. Reply
  • tipoo - Monday, April 07, 2014 - link

    Remarkable how little of the die the CPU cores use. Reply
  • ratte - Monday, April 07, 2014 - link

    From the article: "the die shot above is inaccurate" Reply
  • CiccioB - Monday, April 07, 2014 - link

    Independently of the wrong die shot, CPU part in ARM SoC are only a small part of the entire SoC, unlike on x86 based SoC. Reply
  • extide - Monday, April 07, 2014 - link

    LOL, "unlike on x86 based SoC" ...

    Check out a bay trail SOC -- which would be comparable to this one http://techreport.com/r.x/bay-trail-preview/die-sh... -- looks like all 4 CPU cores only take up about 25% of the whole area. Or how about the PS4 and XBONE SoC's ? Lol...
    Reply
  • Wilco1 - Monday, April 07, 2014 - link

    Look at Atom vs Jaguar or A15: http://chip-architect.com/news/2013_core_sizes_768...

    We don't know the exact die size of BayTrail, but we do know the package size of BT (eg. Z3770) is 51% larger than that of its predecessor Z2760. That suggests BT is still significantly larger than A15 or other ARM cores.
    Reply
  • Speedfriend - Monday, April 07, 2014 - link

    Qualcomm high end 64 bit SOC shipping only by H1 2015, with their proper version by H2 2015? While Intel will be shipping 64bit Android tablet and phones this year. This may be the marketing hook that Intel needs to get into the market properly. Qualcomm must be kicking themsleves. Reply
  • grahaman27 - Monday, April 07, 2014 - link

    Or nvidia. Reply
  • CiccioB - Monday, April 07, 2014 - link

    Marketing apart, what are you using 64 bit on your smarphone/tablet for? 4GB of RAM? Super computing requiring 64bit math 100% of the computing time?
    Intel is not going anywhere with their 64bit Atoms as it went no where with their 32 bit ones. They are simply too big and too expensive.

    64 bit are useful in Server market and when mobile devices will start using more than 4Gb of RAM.
    Reply
  • Speedfriend - Monday, April 07, 2014 - link

    @ Ciccio8 We all know people buy computing on perception of performance and need. Do we need 8 cores, no but many markets demand it. The next step will be 64bit regardless of if it does anything. No-one in a shop knows how Intel's chips measure up against an ARM version, they just hear Intel and 64bit.

    And Intel atom isn't expensive because it is free...
    Reply
  • extide - Monday, April 07, 2014 - link

    Uh, Atom isnt free ... I mean I love Intel just as much as the next guy but come on, at least be honest! Reply
  • CiccioB - Monday, April 07, 2014 - link

    Intel chips cannot be cheaper than ARM based ones because for the former to compete with the latter a more advanced PP is needed and more advanced PP cost more per mm^2.

    ARM based deevices market is a low margin gain market. Only those that sell a lot of devices at cheap prices survive. I can't see Intel in this market as it can't compete price wise (performance is not a issue here neither for ARm which are "fast enough" not for Intel that is "more than faster enough but less energy efficient").
    Remember that no company that makes ARM devices has an advanced semiconductor fab to upgrade every couple of year. They pay 90 for an ARM chip made by TSMC and sell them for 100, making a net gain of 10. Intel already has a cost higher than 90 as for their more advanced PP and has to accumulate more money than those 10 for a new PP to stay ahead of TSMC.
    On the same PP Intel SoCs have no hope to be competitive in any way, that's why Intel fastly brought Atom from being made by 2 older PP to the newest available one in 3 years forgetting about their Tick-Tock strategy used for other architectures where they are actually monopolist.

    Where Windows application compatibility is not needed, Wintel joint-venture is also not needed.
    Mobile market seems to be the strongest enemy of both companies that, look at their numbers, are quite behind all the other actors on that market and there's not really any reason to suppose that will change (it's easier for MS to take a part of the market being ARM compatible and offering services that can make some difference, but for Intel, what for their SoCs that have not proven to be better that ARM based ones for the last 3 years? And next years TSMC will have 16nm FinFet, probably worse than Intel 14nm FinFet but still good enough to keep ARM SoC ahead of any Intel SoC for few years, the time required to Intel to pass to the new PP after 14nm.. if they will really able to afford that).
    Reply
  • dylan522p - Monday, April 07, 2014 - link

    Intel offers Atom at prices about equivilant to Qualcomms offerings. Atom is fast as all of Qualcomms offerings and uses less power. They are behind on the GPU (solved by using the supriir Imagination graphics) and behind on modem. 16nm FinFet is 20nm FinFet but the small guys want to lie and pretend like they are even close to Intel. Reply
  • Wilco1 - Monday, April 07, 2014 - link

    Intel's 22nm has about the same density of 28nm TSMC. Who is lying now? Reply
  • akdj - Monday, April 07, 2014 - link

    Boy, some things 'never change' eh? :)
    We don't need 'quad/64bit/4GB RAM' phones TODAY...But tomorrow? I've got the 5s (iPhone, personal) and the Note 3 (business) handsets. Three times the RAM in the Note3, perceived performance? About the same. The 7.1 update to the iPhone has helped the A7's performance (or challenges of) significantly and as iOS continues 7.1.x optimization...and delivers their SoC this fall, I think it's quite obvious where they'll go. Higher DPI (larger phone) with more, possibly faster (DDR4 this year?) RAM. Not so much a boost with computational power in the A8 vs graphics. As they've done with the nMP,..moving forward, I think it's obvious where the 'horsepower' should be. Within another half decade (today's phones are faster than laptops in 2008 already!)---our 'phone' may actually just be our main computer. Displays where you need 'em. Plug n Play. The power, storage, memory and graphics all provided by the 'phone/tablet' displayed on your new Dynex $699 4k Best Buy Black Friday special;)
    Point being, we DON'T NEED the power today....but in '07(iPhone) and '08 the Androids....that was just five years ago. Even two year differences in power, display technology and the graphic power of these guys.....just look at the Note 3 review and check it's scores in comparison with the Note 1 (hint, Note 1 is usual at the 'bottom' of each 'score sheet' regardless of bench test). So while today might not be the day you're folding DNA sequencing on your phone (while it's charging at night)---2020 may look VERY Different. ESPECIALLY with Intel on board and the phenomenal competition for SoC powered mobile, faster NAND, SSD & PCIe storage, more efficient and quicker memory, quadruple the pixels (4/8k). I'll never forget how excited I was to get my first 1024/768 color display so many years ago...today's technology is mind blowing (I'm 43 & owned an Apple IIe in high school with 5.25" floppies....cell phone wasn't a 'possibility' @ that point;)). My nine year old son has 100, maybe 10,000x that power in his iPod touch or his iPad 2 today. Cool time to be a 'geek' for sure.
    J
    Reply
  • dylan522p - Monday, April 07, 2014 - link

    Atom uses less power and is cheap as Qualcomms current offerings. Reply
  • CiccioB - Tuesday, April 08, 2014 - link

    It can be cheaper as you want, but that just means Intel is loosing money on them. More advanced PP and bigger die mean more cost. Price is another thing.

    About consuming less than an ARM SoC, please, be serious. You can't really believe that all phone/tablets producers just discard Intel Atom "super power and efficient architecture" if that was true.
    Simply, as proven in the past, power measures on Intel devices are a lie. Intel has always tried to compare its performance/power consumption with ARM SoC much older than those present on the market. See the old first Atom vs Cortex A8 when A9 was already well wide spread, see the subsequent comparison with A9 when A15 was already released, see the fact that now all still to be released Intel Atoms compare to A15 when A57 is round the corner and will surely be released earlier than any new Intel 14nm architecture... consider that Intel 14nm architecturew is going to fight against next year 16nm FinFet, which again, will be worse than Intel 14nm FinFet but good enough to put any ARM SoC buint on that PP ahead of any Intel solution.
    Intel is simply a step behind in this market. The rate at which its products are brutally unrecognized as being good is evident. In the past 4 years Intel has sold as many SoC as Qualcomm has in possibily one months. Maybe less. And that only because Intel is subsiding projects and greatly discounting SoCs.

    You can't really believe Intel can sustain an entire business based on more advanced fab with this selling rate at discounted prices.
    Reply
  • Wilco1 - Monday, April 07, 2014 - link

    Mediatek will have several Cortex-A53 and A57 based designs out by Q3 this year. Samsung is also not sitting idle, and although they haven't announced parts, they have publicly stated to have 64-bit SoCs this year (likely also A53/A57). So it seems likely Samsung and Mediatek will gain marketshare on the high end from QC. I don't see Intel gaining share as the mobile version of Silvermont isn't out yet and we already know A57 will be significantly faster. Reply
  • Speedfriend - Monday, April 07, 2014 - link

    @ Wilco1 Do you have a link to A57 versus Bay Trail/Silvermont? Samsung and Mediatek 64bit are likely to only be out towards the end of the year or early 2015 from what I have seen. Reply
  • Wilco1 - Monday, April 07, 2014 - link

    No, however we know A57 is ~50% faster than A15. And we also know BT is already slower than A15, eg. 2.4GHz Z3770 barely outperforms 1.9GHz A15: http://browser.primatelabs.com/geekbench3/compare/... (that's with BT using hardware acceleration for AES).

    Mediatek says Q3 for some of its 64-bit cores, others are Q4. Not sure about Samsung. There is also NVidia of course, several benchmarks of Denver have been leaked, so launch must be near. As for the mobile version of BayTrail, the initial variants are relatively slow dual cores. I don't think they will be able to compete with existing 28nm Exynos, Tegra and Krait quad cores, let alone next-generation 20nm A57. Intel will need a completely new microarchitecture to compete or be relegated to mid and low-end.
    Reply
  • smartypnt4 - Monday, April 07, 2014 - link

    Those are all excellent points, and you may very well be correct in all of them.

    However, Intel's Cherry Trail (new microarchitecture codenamed Airmont) at 14nm is supposed to ship in 2014 for tablets and hit clocks of 2.7GHz (according to very, very sparse rumors). This will be what competes with the quad-A57 designs. No idea on Airmont's performance vs. Silvermont's, though (not even rumors). Intel's Moorefield (dual and quad Bay Trail for phones) will again be relegated to mid-tier status in Q3/Q4 of this year when the A57's launch. So while yes, Intel still will be substantially behind in phones, I don't think they're far behind at all in tablets, provided they can actually launch Cherry Trail in Q4 of this year, and provided Airmont provides a decent uptick in performance. But we shall see, I suppose.
    Reply
  • Wilco1 - Monday, April 07, 2014 - link

    Based on Anand, Cherry Trail is a shrink of Bay Trail to 14nm. Obviously there will be some performance tweaks, and a frequency gain to 2.7GHz seems plausible, but I'd be surprised if maximum performance increases by more than 20% - not enough to compete with Tegra K1, let alone Denver or A57. A Q4 release means it will be in the middle of the big wave of 64-bit ARM chips starting from Q3, so it may be leapfrogged before it is even released...

    2014 is certainly going to be an interesting tech year!

    In any case, it looks like 2014 is going to be a very exciting tech year!
    Reply
  • smartypnt4 - Monday, April 07, 2014 - link

    I was under the impression that Airmont qualified as another "tock" in the roadmap due to Intel wanting to catch up, but I could be wrong there. Most of what I've seen seems to indicate that Airmont is closer to a "tock" than a pure "tick" in Intel's parlance. Then again, maybe I'm thinking of Willow Trail, which is the Goldmont microarchitecture, and is supposed to release somehow within 6-8 months of Cherry Trail. No idea, really. This fall is going to be incredibly jam-packed with stuff, though, what with Apple's new stuff that's rumored to use LPDDR4, Intel's Cherry Trail, Snapdragon 805 (not that interesting on the CPU side, but still), and especially NVIDIA's Project Denver.

    In any case, I'm more interested in seeing how 16 Broadwell EUs compete with the Adreno 420 and Tegra K1 on the GPU side.

    2014 is indeed shaping up to be a very, very interesting year.
    Reply
  • Gondalf - Monday, April 07, 2014 - link

    Goldmond, a quarter after Airmont Reply
  • virtual void - Monday, April 07, 2014 - link

    Most benchmark sucks, especially for mobile units where the ability to react quickly to short burst are way more important for the perceived performance of the SoC than the maximum throughput.

    How do "we" know that A57 is ~50% faster than A15? I can "prove" that going to IA32 to Intel64 yields a 57% on Silvermont: compile this
    http://benchmarksgame.alioth.debian.org/u32/progra...
    program with gcc 4.8.1 with "-O3 -m32" and "-O3 -m64" on Silvermont machine running 64-bit version of Linux. The Intel64 version is 57% faster than the IA32, you can even get an additional 7% speed-up by using x32 ABI ("-O3 -mx32").

    That is clearly not a typical performance increase, but Silvermont (for whatever reason) benefit more from IA32->Intel64 than the "big-core" models do.

    If Cortex A57 is so fantastic, how come that AMD predicts its upcoming Cortex A57 microserver to have a lower SPECint score than Intels Avoton? Both are 8 core parts but Avoton has somewhat lower TDP.
    http://www.anandtech.com/show/7724/it-begins-amd-a...
    Reply
  • Wilco1 - Monday, April 07, 2014 - link

    Anand quoted 25-55% in this article. I expect good gains from 64-bit just like A7 in Anand's review. You're right that Silvermont will likely show a slightly higher gain from 64-bit due to being register starved in 32-bit mode. Although we don't have good comparisons today, it doesn't mean we can't conclude that A57 will be significantly faster given the fact A15 already has significantly better IPC than Silvermont.

    Comparing SPEC is fraught with issues. Intel uses ICC which games several of the SPEC benchmarks resulting in huge speedups. While this results in great SPEC scores, it doesn't translate into real world performance (unless all your code is exactly like libquantum). So if you were hoping Avoton can somehow beat A57 then you're going to be very disappointed.
    Reply
  • virtual void - Tuesday, April 08, 2014 - link

    Ah, so what you claiming is that SPECInt is not relevant because running real programs like gcc, bzip2, perl-interpreter are all things that can be "fooled" by "compiler tricks" while real benchmarks like GeekBench really show the truth?

    You do realize that Geekbench include test like "stream copy" that look like this

    #pragma omp parallel for
    for (j=0; j<STREAM_ARRAY_SIZE; j++)
    c[j] = a[j];

    something that gcc recognize and replace with a call to memcpy()... A little more complicated one is "stream scale"

    #pragma omp parallel for
    for (j=0; j<STREAM_ARRAY_SIZE; j++)
    b[j] = scalar*c[j];

    where one for some reason decide to use floating point arithmetics (which do use x87-ops on IA32 but uses SSE2 on Intel64).

    You can't be serious about Geekbench being more relevant than SPECint... Not that SPECint is very good at predicting how good/bad a CPU for mobile apps.
    Reply
  • Wilco1 - Tuesday, April 08, 2014 - link

    No, I am claiming that SPEC comparisons are more compiler trick contests rather than real CPU performance comparisons. I know as I've been writing compilers and doing benchmarking for decades now.

    If you compare SPEC compiled with the same GCC compiler with the same options used for ARM and Silvermont then you can get a good idea of the performance advantage of A15 or A57. If you use ICC than the 30-40% advantage it has on SPEC can completely hide any CPU performance difference. Then Intel obviously claims it has a huge performance advantage when it is solely the result of special optimizations in their ICC compiler. And as I said those special optimizations ONLY help SPEC, nothing else.

    Geekbench does not suffer from this as it uses GCC and LLVM. Geekbench is not the best benchmark in the world, but at least it tries to give an HONEST indication of CPU performance, unlike what Intel is doing with SPEC.
    Reply
  • virtual void - Tuesday, April 08, 2014 - link

    The difference between ICC and gcc is minimal these days, even clang is very close. There are odd corner cases one could use to show that any of these three compilers is better or worse than the others. You cannot be serious in claiming that ICC can make things like the bzip2, gcc-compiler and perl-interpreter go fast via "compiler tricks", ICC doesn't those program any faster than gcc and if it did, then it would just be a superior compiler as we are talking about _real_ programs that solves _real_ problems.

    And come on, apart from Geekbench, point to any other benchmark/program where Z3770 does not clearly beat Tegra4 in CPU performance. How come that Geekbench, the only thing that seem to deviate, is the "one true benchmark"?

    Ignoring other flaws, Geekbench and SPECint still both fail at being any good at predicting how good a CPU would be for mobile/tablet as they contain too much floating point tests (and Silvermont kind of suck at floating point by design, Intel know it is not important here and don't want to waste space on that).

    I can understand that one would assume that A15 clearly beat Silvermont if one look at the microarchitechture, but people seem to forget/ignore just how much better Intel is at cache-design compared to any other CPU-designer right now. An we all know that cache is king!
    Reply
  • Gondalf - Monday, April 07, 2014 - link

    As usual you are wrong. You have not a single serious bench proving your claims.
    In fact Avoton is faster and cooler of the new Amd 8 core server puppy based on A57.
    You can not judge a cpu with a shitty syntetic bench like GeekIdiocy......the name says all the story.
    Reply
  • Wilco1 - Monday, April 07, 2014 - link

    You're just deluding yourself. None of the Atoms have been successful in any way, Intel is losing billions of dollars a year trying to push them. We've seen it several times before with Intel talking up their benchmark scores only to show mediocre results when a tiny fraction of the promised design wins finally enter the market. What exactly makes you think it's different this time round? Reply
  • Speedfriend - Tuesday, April 08, 2014 - link

    @Wilco
    "No, however we know A57 is ~50% faster than A15" - have you got a link for this? Certainly in browsing performance ARM is talking about almost 50% improvement, but another slide I saw shows an overall score suggesting more like 20%. But anyway, I meeting with ARM management in a few weeks time, so I'll ask them.

    The other thing you are forgetting is that Intel may not be using Atom to compete at the upper end of the tablet market, but Broadwell if they manage to get the level of efficiency improvement that they are talking about. When you have good quality fanless tablets or convertibles running Windows, the purchase decision for corporates will change significantly. as much as I love my iPad Air, it is basically a toy for me. I would dream of actually trying to do work on it
    Reply
  • Wilco1 - Tuesday, April 08, 2014 - link

    ARM should be able to show you benchmark results, including comparisons between Silvermont and A57.

    We're certainly at a performance level where you can use your tablet for real work. I'm using an A15-based Chromebook for benchmarking, open source development and serious build jobs. Obviously it doesn't beat a x86 desktop today, but I bet a lot of developers working on projects targeting ARM will start to use ARM-based computers when much faster 64-bit ARMs are out.
    Reply
  • Krysto - Saturday, April 19, 2014 - link

    Silvermont is a full node and a half ahead of ARM in process technology, and it still barely competes. Not to mention Atom chips still aren't competitive at all on price, and Intel has to lose billions every year subsidizing them to make them look half-attractive to OEMs (which would be fools to fall for Intel's tricks, because as soon as Intel becomes stronger in the market - if ever - they are going to raise prices on them immediately).

    http://www.theverge.com/2014/4/15/5618126/intel-is...
    Reply
  • extide - Monday, April 07, 2014 - link

    Since when do we know that A57 will already be faster than phone-baytrail ? Reply
  • Wilco1 - Monday, April 07, 2014 - link

    Since last year we know the fastest Baytrail has about the same performance as Tegra 4. The phone version will be dual core initially and run at a lower frequency. Tell me, are you betting it will beat any of last years high-end phones? Reply
  • dylan522p - Monday, April 07, 2014 - link

    Which is fine as Airmont will be out this year. Reply
  • Gondalf - Monday, April 07, 2014 - link

    Faster of Silvermont?? you do not know this. Without power constrains Silvermont is clearly faster than A57 in server SOCs. Unfortunately you are an entusiast and you have forgotten the thermals in your equation. A phone/tablet SOC can not draw power over a certain limit, even if A57 is faster than Silvermont (and it is not), it will throttle down all the time, like are doing all the A15 based mobile devices around.
    Intel has the best 64bit core in this moment and it will gain a lot of market share expecially in tablets.....not having integrated LTE. Moreover there is a new a faster core on track for H1 2015, it is named Goldmont, likely a little wider than Airmont.
    One thing is certain, Qualcomm has lost the lead over all competitors
    Reply
  • Wilco1 - Monday, April 07, 2014 - link

    We already know A15 outperforms Silvermont by a good margin. A57 outperforms A15 by a good margin. So your conclusion is that Silvermont is clearly faster than A57? Wow, hear the expert talking. And if it isn't faster then we should just wait for the next one? Only in fan-boi world.

    I agree Intel has one of the best 64-bit cores, but unfortunately for Intel Haswell is both to big and inefficient to be used in phones.
    Reply
  • Mondozai - Monday, April 07, 2014 - link

    Most tablets are still at 2/3 GB and that won't change any drastically in any short amount of time.

    Add to that the fact that most Android apps are also not optimized for 64 bits. Plus, the big story now is smartphone/tablet saturation in the developed markets. The growth markets in developing countries don't need high end. They need smartphones like the Moto G for sub-200 dollars and preferably sub-150 dollars.

    For these reasons, 64 bit this year is still in many ways premature. But Qualcomm moves ahead anyway because 64 bit will matter more in 2015 and onwards. Apple was early, but did it really make that much of a difference? Most I hear from even Apple's fanboys is a lot of complaints about memory leakage and crashes. 64 bit still isn't ready for primetime.

    Also, when was Intel actually relevant in the Android space? They do okay on Windows tablets, but Android is gobbling up everything. It's not at 80+ % and it won't be long before it is at 90+ %.
    Windows on tablets, and especially smartphones, is an afterthought at best.

    And Intel is damned late to the Android party. Qualcomm isn't invincible, but people like you who obsesses over Intel are too narrow-minded in the strict geographical sense.
    The big growth markets are in the mid to low end of the segment and Intel is completely creamed there. Even Qualcomm is struggling, but it is doing a lot better than it used to.

    The major competitor to Qualcomm will be MediaTek, as they slowly start to shift their gears upwards in the value chain as their home base of China(even if they're a Taiwanese company, most of their sales are in China) is starting to get a lot more advanced. They'll be able to push up the value chain in China - they already are - while continue to do what made them so formidable in the budget segment for places like Africa, India, the poorer parts of Latin America and the Middle East as well as central Asia.

    People in the Western world tend to be overtly focused on Western companies. Qualcomm does great in the premium segment and people have been predicting that Intel will crush everything for years on end now. They're still behind and will remain so. The major threat to Qualcomm is that MediaTek/Allwinner and other companies can move up the value chain a lot easier than Qualcomm can move down.
    Reply
  • Speedfriend - Monday, April 07, 2014 - link

    @Mondozai - I certainly don't obsess over Intel and am purely interested from an investment point of view. Which means I try not to get bogged down in the tech when it doesn't matter and look at how consumers react.

    Mediatek has been very successful because not only are they cheap but they play the marketing game. They have been the prime drivers of the core race in smartphones. They may be able to push up into high value end, but I'll believe it when I see it.

    As for 'Windows on tablets, and especially smartphones, is an afterthought at best.' I agree on smartphones, although in Italy and Spain, Windows Phone has overtaken IOS. On tablets, I think we are just at the beginning of the potential for Windows tablets. I work in a company where most people are obsessed with Apple products, we use iPhones as work phone, people bring iPads into meetings, but now we are offering tablets as laptop alternatives, we are going Windows 8 because we simply cannot use an iPad as a laptop replacement. If Win 8 tablets take off in the business market, then watch what happens in the consumer market as sub $200 Win 8 tablets launch.
    Reply
  • Krysto - Saturday, April 19, 2014 - link

    You're looking at this the wrong way. It's going to take years to move everyone off to ARMv8 anyway, because of the buying cycles and the product life cycles. So the sooner we start with ARMv8, the better. Also ARMv8 brings other big performance improvements. It's not just the 64-bit support that matters. Reply
  • wwwcd - Monday, April 07, 2014 - link

    with 810 we have supercomputer inside a smartpfone box. Reply
  • grahaman27 - Monday, April 07, 2014 - link

    Hmm... Not sure if serious... Reply
  • Phill49 - Monday, April 07, 2014 - link

    Shouldn't the memory interfaces be 1x 64bit if there all 64bit architectures rather than 2x 32bit? Reply
  • Exophase - Monday, April 07, 2014 - link

    Most of the memory traffic from the CPU comes from cache line transfers, not individual register transfers, so nothing has changed there. A 2x32-bit interface is still more flexible, and could be better for sharing bandwidth between CPU and GPU.

    The 64-bit chips will still mostly run 32-bit software for a long time anyway, and even 64-bit software will use a lot of 32-bit datatypes. I don't know what direction Android is taking with 64-bit ARM exactly but it'd be great if there's a mode that's like x86-32, where user space programs can use the 64-bit ISA and a more open address space while still being restricted to 32-bit pointers for most things. That'd still be sufficient for most programs, while avoiding the data bloat of needless 64-bit pointers.
    Reply
  • extide - Monday, April 07, 2014 - link

    There is a 32-bit mode in ArmV8 Reply
  • Exophase - Monday, April 07, 2014 - link

    But that's not what I'm talking about. This isn't about CPU modes, it's about software modes (enforced by ABIs/compilers/standard libraries/OSes). I'm talking about using the 64-bit CPU mode but not using 64-bit pointers in normal user mode scenarios. There are two advantages: one, you get access to the newer instruction set with more registers and 64-bit arithmetic and various other tweaks. Two, you get an entire 4GB address space instead of 3GB or worse, since the kernel would still use 64-bit pointers and live above the 32-bit address space.

    x86-32 is exactly this, the 64-bit instruction mode but restricted to using 32-bit pointers. 64-bit pointers are a big waste if you don't need them.
    Reply
  • Wilco1 - Monday, April 07, 2014 - link

    Memory interface width has absolutely nothing to do with the architecture bitness (which is just the integer register width). CPUs have been using 64-bit doubles and 128-bit SIMD vectors for years. Reply
  • Aenean144 - Monday, April 07, 2014 - link

    I'm just happy to see "20 nm" on an SoC roadmap. Haven't paid attention to Nvidia's or AMD's discrete GPU roadmap, or AMD's CPU roadmap, but with Qualcomm putting 20 nm out there for these SoCs, give me some confidence that it is coming sooner rather than later, though it feels that it's late already. Reply
  • mrdude - Monday, April 07, 2014 - link

    I can't be the only one disappointed by the chart above. It's not because it's a poor looking chart or anything, but because it showcases how practicality is thrown out the window when it comes to SoC design for mobile phones.

    I find the lack of a dual core A53 option with integrated Cat7 radio, great WiFi, with LPDDR4 built upon 20nm to be really disappointing. In my opinion, that would make for a perfect smartphone SoC, offering enough performance for the form factor yet with great power saving features and battery life. Instead we get chips on a lagging node (lower cost silicon for lower cost chips), quad and 8-core designs for operating systems that barely use two simultaneously and second-rate radio.

    Smartphones are already fast enough, can we please stop trying to make them faster? And the cores aren't helping anything but their bottom line.
    Reply
  • jeffkibuule - Monday, April 07, 2014 - link

    A basic principle of CPU design is that if you can run it faster with only a minimal increase in power requirements, overall you gain more battery life (assuming the task needed to be done has a defined set, the CPU isn't running for infinity).

    I do agree however that on the outside it seems Qualcomm is getting a bit core heavy, when you have to wonder which tasks can be subdivided adequately enough such that power savings can be realized. If a task must be single threaded on a single wimpy core out of 16 or so (taking the desire for more cores to the extreme). It's not going to be better than a single core of a larger dual core CPU.
    Reply
  • mrdude - Monday, April 07, 2014 - link

    It's not just the core count, but the neutering of features and necessitating that only the highest ASP chips -- which would fit a 10" tablet better than a smartphone -- utilize the newest node available (timing and staggered release explain this), the best radio and WiFi. As a result, if you want the best Cat7 LTE you also need to buy the largest and most power-hungry SoC with cores that you won't ever use.

    If these SoC makers spent half as much money on the Android platform and pushing forward a more comprehensive approach to multi-threading as they do marketing their useless core counts, we probably wouldn't have a reason to complain about wasted silicon.
    Reply
  • CSMR - Monday, April 07, 2014 - link

    I hope that these are the chips put out for marketing purposes and there will be more sensible designs in future.

    The megapixel wars and core wars, an the 64-bit and 4k bandwagons appeal to a lot of stupidity in the press and in buyers and companies follow that.

    I like the 20nm and H265 support but 64 bit is irrelevant for phones and going beyond 4 cores when the 3rd and 4th are already useless is crazy.
    Reply
  • jeffkibuule - Tuesday, April 08, 2014 - link

    64-bit ARM comes with an updated ISA which results in improved performance. This isn't just like x86-64 where a few registers were tacked on. Reply
  • chucknelson - Monday, April 07, 2014 - link

    > "I don't know of any other silicon player that can move and ship this quickly."

    This statement is a bit too bold I'd say. While they seem to be "moving quickly", it remains to be seen if they can actually ship any of these on time.
    Reply
  • ahomad - Monday, April 07, 2014 - link

    I don't think you made the calculation right. if the adreno 420 is 40% faster than 330 and 430 is 20% faster than 420 then if we assumed 330 = 100, then 420= 140 and 430=168. so the adreno 430 is less than 70% more power than adreno 330 not more than 80% you mentioned.

    Anyway, this is the biggest disappointment in this otherwise excellent chip. tegra 4 GPU is in the same level as adreno 330, however, tegra 5 is 300% more powerful than tegra 4 while adreno 430 in only 70% more powerful than adreno 330, so tegra 5 is almost twice as (1.8X to be exact) powerful as adreno 430 and at the same time tegra 5 is supposed to be released ~6 months earlier than the new adreno, they should reconsider their chip if they want to dominate the market again especially that 2k displays are around the corner and there is a need for more powerful GPUs
    Reply
  • darkich - Monday, April 07, 2014 - link

    GPU in the Tegra 4 ~ 98GFLOPS
    Adreno 330~115-148 GFLOPS

    Also;
    Snapdragon 805(Adreno 420)~ 25GB/s
    Tegra K1 ~ 17GB/s

    Also;
    Snapdragon 805~1W TDP
    Tegra K1 ~ 3W TDP

    So, only Maxwell based Tegra should be comparable to Snapdragon 810, assuming that they manage to lower the voltage on the same level..but even then, the GPU *should* be in entirely other league than even the Adreno 430, so in a way I agree with you.
    Reply
  • ahomad - Monday, April 07, 2014 - link

    I completely agree with you regarding the battery consumption, but I was comparing the raw power. The GFLOPS isn't a perfect indicators of the actual raw power of the chip (it is one of the best though). but what I meant from my post is that Qualcome need to further improve their GPU. in the past, the jump from 2xx to 3xx was quite huge (over 3X if I remembered correctly) while the jump from 3xx to 4xx is not that big (less than 2X) Reply
  • ahomad - Monday, April 07, 2014 - link

    you can argue that the market is starting to mature and we shouldn't expect 3X improvement every year (although the 3xx stayed almost 2years). However, if they want to keep donating the market, they should do something especially that as I said 2k displays is around the corner. if we compared the GPU vs resolution (ignoring all other factors) fullHD device with adreno 330 should perform better than a device with 2k screen and adreno 430 which is not very good. Reply
  • tuxRoller - Monday, April 07, 2014 - link

    It's actually over 80% because you mixed up the numbers a bit. The 418 is 20% faster than the 330. The 430 is 30% faster than the 420, not 20%. So, is about 82% faster than the 330, which is decently faster than the T4.
    Also, if darkich is right any the power draw that not saying much about the efficiency of the k1. I suppose that's the problem when you have to design a single arch that scales over two orders of magnitude. You have to make compromises at the bottom.
    Reply
  • dabotsonline - Monday, April 07, 2014 - link

    "Only the Snapdragon 810 has a hardware HEVC encoder however."
    This is good. Unfortunately, the Qualcomm press release ( http://www.qualcomm.com/media/releases/2014/04/07/... ) states, "... 4K video at 30 frames per second... The combined 14-bit dual Image Signal Processors (ISPs) are capable of supporting 1.2GP/s throughput and image sensors up to 55MP." This is a shame: I was hoping for 4K HEVC recording at 60fps.

    "The 810 can support up to two 4Kx2K displays (1 x 60Hz + 1 x 30Hz)... "
    This is good. However, the press release states, "... external 4K display support via HDMI1.4." How can HDMI 1.4 support 4K at 60Hz?

    Still, this is not good enough from Qualcomm. By the time the Snapdragon 615 and 32-bit Snapdragon 805 - let alone Snapdragon 810 - are released, Apple will have released the A8. By the time Qualcomm release their Krait-based 64-bit Snapdragon, Apple will have released the A9.
    Reply
  • darkich - Monday, April 07, 2014 - link

    GPU in the Tegra 4 ~ 98GFLOPS
    Adreno 330~115-148 GFLOPS

    Also;
    Snapdragon 805(Adreno 420)~ 25GB/s
    Tegra K1 ~ 17GB/s

    Also;
    Snapdragon 805~1W TDP
    Tegra K1 ~ 3W TDP

    So, only Maxwell based Tegra should be comparable to Snapdragon 810, assuming that they manage to lower the voltage on the same level..but even then, the GPU *should* be in entirely other league than even the Adreno 430, so in a way I agree with you
    Reply
  • darkich - Monday, April 07, 2014 - link

    oops, hit the wrong field, disregard the comment Reply
  • twotwotwo - Monday, April 07, 2014 - link

    Yeep, so if an original 64-bit core is coming from Qualcomm, it's not coming until next gen. They weren't kidding about getting caught off-guard by the move to 64-bit.

    On the other hand, this should be a nice bump, and sounds like folks in the US will finally get to see high-end ARM IP in phones. :)
    Reply
  • Laxaa - Monday, April 07, 2014 - link

    So, get a new phone with 800/01/05 this year or wait for 810? Reply
  • bleh0 - Monday, April 07, 2014 - link

    might as well wait for 810 unless you need a new phone within the next few months. Reply
  • Laxaa - Tuesday, April 08, 2014 - link

    I want a new phone, but I don't really need one. It's not like my 920 is broken. It still functions as well as it did when I bought it. Reply
  • sherlockwing - Monday, April 07, 2014 - link

    They been there before: Nexus 7 and Nexus 7 (2013). So I expect to see Nexus 5(2014) this year. Reply
  • Krysto - Tuesday, April 08, 2014 - link

    Disappointing, to say the least, that they are being forced to use stock designs even though they've licensed the architecture years ago.

    It usually takes 2 years for a new core to appear (H2 2015 at the earliest), which means that if it's taking Qualcomm 2 years since Apple launched their ARMv8 core, to launch their own custom ARMv8 core, then they must've scrapped almost everything they had been working on for their custom ARMv8 core, and started from scratch.

    Also, their GPU performance for 810 seems pathetic. It will probably be 70 percent of mobile Kepler...but in 2015.
    Reply
  • xilience - Tuesday, April 08, 2014 - link

    @anand: Would it be possible to get a comparison article between the current 800/801 released architecture, the upcoming 805 architecture (likely slated for us in LG G3/2014 Nexus 5/Sony Z3), and this newly announced 808/810?

    ie. As a consumer who upgrades yearly, is 805 worth the wait now that 801 is out? It seems 808/810 will offer a big perf/watt jump due to 20nm.
    Reply
  • Fidelator - Tuesday, April 08, 2014 - link

    This is a shame, the 805 should have been the High end SoC for the first half of this year while we should be seeing these by the end of the year, given the delays it is very likely we'll see these by the second half of 2015. Reply
  • iwodo - Wednesday, April 09, 2014 - link

    Lots of FUD or misunderstanding here.
    A15 is NOT faster then Silvermount. Arch Vs Arch wise. And please dont use those stupid browser of Geekbench benchmarks.
    A57 will properly be faster then Silvermont, depending on scenario.
    So let's just say A57 == Airmont
    And Goldmont, should therefore be faster.

    It isn't fair when these comparison are being made in context of Power usage. Arm will definitely have the lead in the low end power usage. But things are definitely way better on the Intel side.

    And Apple SoC is ... just a different kind of beast.
    Reply
  • kingblade20 - Wednesday, April 09, 2014 - link

    Ok, what about the "better cores, not more cores" video from qualcomm?
    can anybody explain to me if we could really get true performance gains for using Octas against quads?
    hopefully someone could help me with this. thanks.
    Reply
  • ClevelandRPenman - Wednesday, April 09, 2014 - link

    The shocking impact is to see Qualcomm not ready for the new 64 bit step after having dominated the 32bit market for some years now. And it is not only Apple (which races in its own market) their problem: nvidia and Samsung already announced their 64 bit architetures while Mediatek and other chinese/tawainese producers already have 64 bit SoC ready. http://qr.net/stx3 Reply
  • RonaldACoady - Friday, April 11, 2014 - link

    One of the biggest benefits the ARM ecosystem has is exactly that point: being able to react and adapt quickly by utilizing 'stock' designs with a significantly lower time-to-market. This model extends to the fab side as well. http://s6x.it/l521 Reply
  • vcarvega - Saturday, July 05, 2014 - link

    Given that this chip won't even release until the second half of next year, has Qualcomm fallen behind Nvidia? I have no idea how the two will compare in terms of performance, but Nvidia will likely have a second gen 64 bit chip out before this one even his the market. Reply
  • AllanMoore - Saturday, December 13, 2014 - link

    I can't even care about 1440p on phones. 1080p @ 5"-5.5", anybody that says they can tell the difference between 1440p and 1080p at that size has either eagle vision or is holding the phone so close to their face that its an artificial use case. - http://picoolio.net/image/e4L Reply

Log in

Don't have an account? Sign up now