30
Oct
image

At ARM TechCon today, the titular purveyor of semiconductors announced its Cortex-A50 series, dubbed "the world's most energy-efficient 64-bit processors." Based on the ARMv8 architecture, the line will launch with the Cortex-A53 and A57 processors, allowing not only for significantly more energy-efficient processing, but SoC scalability that makes the line applicable to devices from smartphones to high-performance servers. The A57 is geared toward high-performance, while the A53 is lauded by ARM as its most power-efficient. Both chips also support 32-bit and 64-bit ARM code, and according to ARM, the A53 can live up to the performance of the Cortex A9 at 60% the die area.

Screen Shot 2012-10-30 at 12.22.41 PM_575px

According to AnandTech, both the A57 and A53 can be used on their own, or in ARM's big.LITTLE configuration in which big, powerful cores are grouped with lower performance cores on a single chip. This is the most likely configuration we'd see the cores take in a mobile application (though the A57 on its own can run in a server environment on up to 16 cores).

Using a chip with two A57s in tandem with two A53s for example, a device's OS could switch between the cores and groupings depending on what the device is doing. The A57s would theoretically step up to the plate during high-performance gaming and web browsing, while basic actions like UI navigation would fall back on the A53s. As Anandtech Notes, this four-core setup would actually be classified as a dual-core processor. The configuration is similar to NVIDIA's approach with the Tegra 3 set.

Screen Shot 2012-10-30 at 12.22.25 PM_575px arm

So far, Cortex-A50 licensees include such notable names as AMD and Samsung, and all licensees can expect to start getting completed designs by the middle of next year with actual production to follow by late 2013. The first designs will of course shave nanometers off the previous line, set to ring in at 28nm and 20nm, with a trajectory for thinner goals.

Of course, we're a long way off from seeing the Cortex-A50 line integrated with our handsets, but these chips promise to – as always – push our mobile toys to be thinner, faster, and more energy efficient. We'll keep you up to date on further developments. In the meantime, get your fill of information on the subject from the links below.

Source: ARM via Anandtech

Liam Spradlin
Liam loves Android, design, user experience, and travel. He doesn't love ill-proportioned letter forms, advertisements made entirely of stock photography, and writing biographical snippets.

  • http://twitter.com/TheGermian Roberto Giunta

    So, theoretically in 2 years I can get a PC that is powered by this CPU and run Windows 8 (not RT) on it without a customized/tailored OS?

    • http://www.bradley.zxq.net/gallery Brad Biederman

      This is an ARM cpu, not x86. You could only run Windows RT with this.

    • Cenarl

      Well AMD chips paired with this would be your answer as that would be the only way they could run legacy x86 Applications and APIs, unless an emulator would be legal. That's just a guess because Intel has very strict x86 license agreements with AMD and someone like Andantech will probably elaborate this much better at some point

      • Jay T

        What?? Intel doesn't own x86, the way ARM owns the ARM architecture! AMD makes it's own processors, and isn't licensed from Intel.

        • Cenarl

          AMD does make their own processors but the only reason they can be used on Windows is because they license x86 from intel. Intel had to do this as to not be a monopoly. No one else has an x86 license besides AMD and AMD is not allowed to license it to anyone else.

          http://en.wikipedia.org/wiki/X86

          this is why the ARM powered Windows surface RT cannot use any x86 windows applications...everything has to be created like it was a new OS.

          • ssj4Gogeta

            VIA also have an x86 license. There Nano processors are x86.

          • Jay T

            Oh, right. Hmmm, I'll need to read up on that. You learn something new every day.

          • DCMAKER

            you got a link to that. i couldn't find it after 10 mins and i know there are about 20 companies in the past that have built x86 CPUs. IBM,Fujitsu and many others

          • DCMAKER

            found this in wikipedia. You are wrong Cenarl it is partly open. AMD and Intel own various IPs on each part of the technology. So Intel has to license from AMD and vice versa depending on what is used.

            "Partly. For some advanced features, x86 may require license from Intel; x86-64 may require an additional license from AMD. The 80486 processor has been on the market for more than 20 years [1] and so cannot be subject to patent claims. This subset of the x86 architecture is therefore fully open."

          • ssj4Gogeta

            So what you'll be able to build is a 486 compatible processor, which will be useless without all the x86 extensions, as most modern software won't run on it. AMD and Intel have a cross-licensing agreement that allows one to use any extensions that the other develops.

          • DCMAKER

            it is too vague to really tell by that description what is actually patented. I doubt it would be too hard to build an x86 CPU with or without licenses.

          • ssj4Gogeta

            The extensions are recent so their patents haven't expired. They're added every new microarchitecture. If anyone could build a compatible x86 and legally sell it, they would, because the market is so huge.

          • Cenarl

            if x86 was open in any meaningful way Windows Surface RT wouldn't exist. ssj4 pretty much covered that point though.

          • rstat1

            x86 has nothing to do with Windows. Windows is written to work on x86 chips. Not the other way around.

          • Cenarl

            herp derp, up until now MS never fooled with another processor so x86 has EVERYTHING to do with Windows and legacy apps. Go ahead and build your next Windows PC with an IBM chip and tell me how that works out.

            Just like when Apple made the switch from IBM to Intel...their Desktop OS is still their OS but they made it rely on intel x86 processors going forward so now thats what they are stuck with unless they want to change again.

            they are separate, but forever intertwined.

          • rstat1

            Still wrong. Back when the DEC Alpha and PowerPC chips were actually popular there were versions of Windows NT that ran on them. This stopped with Windows 2000. Also, a bit more recently Windows was built for the Itanium architecture as well, though I think that stopped with Windows Server 2003. There also exists a canceled port to Sun's SPARC architecture So nice try but you're still wrong.

            To say that Windows and x86 are intertwined shows a fundamental lack of knowledge regarding the design of NT kernel. Hint: It was designed to not be dependent on any architecture, but instead dependent on the many different abstraction layers MS has built over the years.

          • Cenarl

            You are scraping the bottom of the barrel. You might as well throw in past windows mobile OS too if you want win your argument with niche markets.

          • rstat1

            Wasn't aware WIndows and Windows CE (ie WIndows Mobile) were even remotely the same thing. Oh wait...they aren't. The only similarity between them is that they share a name. So why would I count that? Also you made the bold claim that Windows only ran on x86...which is wrong no matter how widely used the other ports were.

    • Jay T

      I dream of the day I can run all my apps on an Android tablet hybrid..... Alternatively, I would accept a Ubuntu on Android computer. :D

  • btod

    I'm looking forward to the day a 1000 mAh battery will be able to last me a several days.

    • ssj4Gogeta

      Or the day when we can fit 10x that capacity in a battery of the same mass and volume.

      • Matthew Fry

        Much less likely to happen. Battery technology has stagnated.

      • DCMAKER

        well nanowire batteries were due out already and i am still waiting....so much for 8x capacity being out already :/

        For all you who don't keep up with tech read this

        http://en.wikipedia.org/wiki/Nanowire_battery

        Though I could have sworn he talked about nanowire batteries in 2004 and they were supposed to be out in 2009 but all i know it was a long time ago and i am still waiting

        • ssj4Gogeta

          I heard about it. But what I want is antimatter batteries. :P

          • DCMAKER

            never be that small lol.....those would only work in our star ships :)

  • mechapathy

    So how long til we have phones that have >4GB RAM?

    • Cenarl

      they could do it anytime if they wanted, but my guess is galaxy s5 2014 or note 3 late 2013

      edit: well my estimation is more of an "= 4GB" then greater than

      • mechapathy

        Nvm, I guess ARM already cracked the 4GB limit using virtual addresses.

        • defred34

          I think we'll stay at 2GB for another 16 months.

      • DCMAKER

        first armv8 can be 4GB plus so i would expect the highend phones to have that much or close to it at that time. It really depends if win8 RT takes off. They won't through in 4GB if their is no real gain. I don't see how android can use that much in a phone or tablet. Only time i use that much RAM is running opera...largely because i always have 4 windows with anywhere from 40-80 tabs open lol. Opera uses around 1-3GB on its own :/

  • ssj4Gogeta

    Very interesting to see AMD as a licensee. That might give ARM the headway it needs in the server space and keep Intel on their toes.

    • DCMAKER

      Have you seen Intel's MICs? aka knight corner? It is pulling 1TFLOP with 62w TDP lol Nothing beats that right now. Not even GPU super computers. Plus Intel is going to 14 nm in just over a year from now so knights corner will be revamped around that time too so i highly doubt ARM is going to make Intel sweat much.

      • ssj4Gogeta

        Are you sure it has a TDP of 62W? I read it used around 250W. It descended from Larrabee and a part of the reason why they scrapped it was power consumption. It's amazing if they've been able to bring it down to 62W.

        • DCMAKER

          on the 22nm process it is 1 TFLOP for 62w TDP. It runs at 3GHz. They did try running it at 5.x GHz but that used 250w TDP and only puts out ~1.8 TFLOPs so it proved to be best to have many cards at lower freqs. I am amazed at the efficiency and how much power is required to scale up lol. I would love to see a full blown test seeing performance per watt from 1GHz all the way to 5GHz to see scaling.

          • ssj4Gogeta

            I still don't believe the 62W figure. Can you give me sources?

            Semiaccurate says it's ~200W: http://semiaccurate.com/2012/09/14/hard-numbers-for-knights-corner-leak-out/

          • DCMAKER

            http://en.wikipedia.org/wiki/Intel_MIC

            http://www.xbitlabs.com/news/cpu/display/20070212224710.html

            official intel document
            http://download.intel.com/pressroom/kits/Teraflops/Teraflops_Research_Chip_Overview.pdf

            unsure of what the nm is on this chip but i think with this prototype it is 22nm but i could be wrong. Documentation on this is very vague still. I don't even think there is more than 3-5 documents released by Intel yet. I know i have read 2 intel documents and 1 PPT from intel from all of my googling so info is extremely limited.

          • ssj4Gogeta

            Sorry, but you're confusing two entirely different projects. Polaris and Knights Corner. Polaris was an 80-core research chip which (as far as I remember) wasn't x86. It was just a bunch of simple cores designed for throughput performance. From what I remember, it was shown around 2006 and was built on a 65nm process. What was interesting was the fast terabit interconnects used for connecting the cores. This article is about Polaris: http://en.wikipedia.org/wiki/Teraflops_Research_Chip
            The 3GHz and 5GHz which you mentioned are also there, and they seem to be for Polaris as well.

            Larrabee came after that. It was x86, presumably with some extensions. Prototypes were 45nm. It was too power-hungry and didn't perform up to expectations. They were going to do graphics in software on that chip as well as raytracing. It was scrapped and its next iteration (or next-to-next) is Knights Corner and it's a 22nm chip. And everywhere I've read about it says it has a TDP of 250W. It's definitely not 62W.

            Nvidia Tesla (which is a direct competitor) retails for ~$3000, so I'd expect an MIC card to be around the same.

          • DCMAKER

            i don't know where your going but knights corner is a 50 core MIC and it does 1 TFLOP and i could have sworn it was 62 watt.

            http://www.tomshardware.com/news/intel-knights-corner-mic-co-processor,14002.html

            Also knights corner is basically a revamp of all those dead ends...aka polaris and larrabee.

            If polaris/terascale is doing 1TFLOP at 62w as a research chip at 45 or 65 nm than knights corner with 50 cores and on a 22nm process will be way better.

            http://download.intel.com/pressroom/kits/Teraflops/Teraflops_Research_Chip_Overview.pdf

            I can't find anymore information than these and i honestly don't care its 5:22am and i am going to bed. Either way knights corner does 1 TFLOP and if terascale was doing 1 TFLOP with 80 cores at 62w than knights corner has to be better...no way it is going backwards. Also the 250w can refer to either terascale running at 1.8TFLOPs or knights ferry or larrabee...which ever one it was.

          • DCMAKER

            Also from what i have read and understand it went something a long the lines

            larrabee
            polaris???/terascale
            knights ferry
            knights corner which will actually hit market in the next few months i think
            and second gen knights landing

            Also read this in wiki and i think this best describes the confusion

            "The Teraflops Research Chip (prototype unveiled 2007[7]) was an experimental 80 core chip with two floating point units per core implementing a 96-bit VLIW architecture. The project investigated intercore communication methods, per-chip power management, and achieved 1.01 TFLOPS at 3.16 GHz consuming 62 W of power"

            again if a research chip did this 5 years ago i can't see how knights corner can be slower and more power hungry....just doesn't make sense

          • DCMAKER

            BTW i have rechecked everything and i don't see a single reference to TDP for knights corner only knights ferry and terascale/polaris

          • ssj4Gogeta

            "Either way knights corner does 1 TFLOP and if terascale was doing 1 TFLOP with 80 cores at 62w than knights corner has to be better...no way it is going backwards."

            You don't understand. They were different microarchitectures. That changes everything. One is x86 and has more complex pipelines, the other doesn't. Just like how a 6870 GPU can achieve 2TFLOP @ 150W, while a Sandy Bridge CPU can only achieve ~100GFLOP @ 100W.

            Also, Larrabee came after Polaris and there are many references saying it has a TDP of 250W. Either way, we'll know in a couple of months.

          • DCMAKER

            polaris/terascale says according to intel docs 1 TFLOP at 62w and 1.8 TFLOPs at 250w....read the intel doc. It was an 80 core research chip in the terascale research division....according to you that is called polaris. I don't care what references you have if the intel doc says terascale is 1TFLOP at 62 watts than that is what it is

          • ssj4Gogeta

            Let's not be too hostile here :)

            And I didn't know 1TFLOP was for double-precision. But anyways, I'm not contesting the fact that Polaris does 1TFLOP at 62W, I'm saying Knights Corner's TDP is 250W.

            Don't get me wrong, it'd be awesome if that thing has such a low TDP. But here are my sources:
            The image shows what is presumably an official Intel utility for monitoring the cards. Card 1 shows 190W, Card 2 shows 248W.
            (Image attached courtesy of SemiAccurate.com)

            More sources: http://semiaccurate.com/2012/09/14/hard-numbers-for-knights-corner-leak-out/

            Quote from Anandtech:

            Superserver 2027GR-TRF which contain 4 Xeon Phi cards thanks to two redundant 1800W (Platinum) PSUs. The rest of the server consists of two Xeon E5 and 16 DIMM slots in total, supporting up to 256 GB. So it seems that one Xeon Phi card consume about 300W.

            http://www.anandtech.com/show/6265/intels-xeon-phi-in-10-petaflops-supercomputer

            Quote from cpu-world:

            The 57C/3GB and 61C/6GB models will have a TDP of 245W, while the 57C/6GB and both the 61C/8GBb models will have a TDP of 300W.

            http://www.cpu-world.com/news_2012/2012080201_Details_of_Intel_Xeon_Phi_coprocessors.html

          • DCMAKER

            BTW I know the difference in x86/ARM/GPU but you are also comparing apples to oranges. 6870 is single precision and Intel linepak does double precision

          • DCMAKER

            P.S. whenever i build a desktop i am totally getting 1-2 of these for F@H :)

  • Jay T

    I'm sorry, but I couldn't resist pointing this out: You classed web browsing and gaming as high performance tasks, and browsing the UI as low performance. Actually, web browsing is like walking, requires almost no power and can be done all day. The only thing that even strains the processors these days is getting super smooth scrolling and zooming, but these are only done about 5% of the time max.

    However, Android has a pretty graphics intense UI. Fade and shrink effects? Semi-transparent notification pane? Everything is GPU accelerated.
    :) Looking forward to the new processors, but because I want better battery life.

    • Shhh

      Web browsing is very resource intensive due to PARSING. It's ok, you're obviously not a coder.

      • Jay T

        Hehe. Parsing isn't very resource intensive. You can judge when it stops really accurately, by when you can SEE the page. By that time, the vast majority of parsing is done. The thing that actually can continue to use the processor a lot is javascript, which can keep running. However, most sites are well enough designed that the javascript doesn't stress your phone/tablet at all.

        Also, fun fact. I am a coder. I use C and Javascript. :D

      • rstat1

        Parsing is a VERY VERY small part of the rendering of a web page. If anything, the most intensive parts are the actual drawing and JS. It's ok, you're obviously not a coder.

  • defred34

    Wait, so that means the dual-core A15s are going to be the best till the A57s come out in late 2013/early 2014? That would make it a great thing to invest in a Nexus 10 (considering there won't be quad-core A15s).

    • Cenarl

      Quad cores are coming, this is already known. Samsung will have one, Tegra 4.

      http://lmgtfy.com/?q=quad+core+a15

      They'll probably put on in the SGS4 if i had to guess.

  • dobvlr

    That is a super comupter spec when 1990s-era.

  • Bleakvision

    Can't wait for 4GB of RAM:)

  • Matthew Fry

    What happened to the A15 which hasn't actually been seen yet outside of Qualcomm's implementation? Are they abandoning it?

    • DCMAKER

      Samsung is making their own a long with other manufactures. The A15 are coming out now. Not sure if one is already out or not....don't keep up with ARM as much as I do with x86.

    • ssj4Gogeta

      The Nexus 10 and the new Chromebook from Samsung use dual-A15 (Exynos 5250).