ARM intros 64-bit Cortex-A50, to be used by AMD

Yesterday, we reported that AMD’s ARM-compatible processors will include pre-fab 64-bit ARM cores, not compatible cores designed by AMD itself. Now, we have some idea of exactly what cores AMD is going to use.

ARM has announced a new family of processor core IP, the Cortex-A50 series, which is based on the 64-bit ARMv8 architecture. ARM says the first chips based on the A50 series will ship in 2014. Guess who’s in the list of "announced licensees" for the new cores? That’s right, AMD. (Other licensees include Broadcom, Calxeda, HiSilicon, Samsung, and STMicro.)

The Cortex-A50 series will initially be made up of two models, the Cortex-A57 and Cortex-A53. Here’s how ARM describes them:

The Cortex-A57 is ARM’s most advanced high-performance applications processor, while the Cortex-A53 is the most power-efficient ARM application processor. The Cortex-A53 is also the world’s smallest 64-bit processor. They can operate independently or be combined into an ARM big.LITTLE™ processor configuration, combining high performance with power efficiency. Both are supported by the ARM CoreLink™ 400 and new CoreLink 500 series system IP fabric solutions.
ARM partners can scale SoC platforms from single- and multi-core big.LITTLE mobile solutions to massively parallel enterprise solutions for optimal flexibility and energy-efficiency. The Cortex-A57 and Cortex-A53 processors will target multi-GHz performance on advanced CMOS and FinFET processes technologies, which is supported by early availability of ARM Artisan® Physical IP and ARM POP™ IP for core-hardening acceleration.

Here are some simplified architectural diagrams showing the two offerings. Not pictured: ARM’s new terabit CoreLink CCN-504 interconnect, which will be supported by both designs.

According to ARM, the Cortex-A57 will deliver the "highest single-thread performance," and systems featuring it will have performance "comparable to a legacy PC." Also, as the company notes in the quotation above, the Cortex-A53 and Cortex-A57 will both support both CMOS and FinFET fab processes. (FinFET is shorthand for three-dimensional transistors.)

GlobalFoundries, which produces AMD’s processors, partnered with ARM to promote the migration to FinFET devices this summer. A month later, GloFo announced that its 14-nm FinFET process will be ready in 2013. Perhaps the first processors based on the Cortex-A50 series will use that process.

Comments closed
    • Arag0n
    • 7 years ago

    Windows RT 64bits with a proper metro ecosystem can kick Intel out of the market…. ARM has an open licensing scheme that makes CPU development inexpensive. Windows RT with a 4 core A57 CPU plus an integrated GPU can be a very nice entry level computer for so many people. Most of the missing productivity apps can be developed for Metro, there is no limitation about why Adobe can’t have a metro version of photoshop.

      • shank15217
      • 7 years ago

      you’re right, most users want to buy hardware slower than their 2008 pcs that doesn’t run any software built before 2012. Intel can take on arm alone and make no mistake if amd leaves the x86 space they will die for good. Its always harder to compete in a market with several players.

        • Arag0n
        • 7 years ago

        If they can get 25% of Core i5 performance and better than Ultra Book battery life for 5$ sure they will.

    • Essence
    • 7 years ago

    I cant wait for the AMD APU (ARM+Radeon GPU). Many ATi Engineers still at AMD have experience in ARM including Jim Keller.

    The ARM APU is going to be AMD`s entry into low power mobile stuff.. I just hope it comes before end of 2014 after the server parts are released.

      • TaBoVilla
      • 7 years ago

      AMD sold it’s handheld graphics division to Qualcomm back in 2009, including intellectual property, patents and stuff created by ATI back then, which set the foundation for the Adreno line of mobile graphic chips (Adreno being an anagram for Radeon) currently used by Qualcomm Krait chips, like their latest Adreno 320.

      It’s unlikely they (AMD) would be able to use any IP it had, so without licensing they basically have to “reinvent” their mobile GPU architecture, which would be quite difficult and expensive.

        • Helmore
        • 7 years ago

        Was the deal that Qualcomm struck with AMD really one where Qualcomm got exclusive rights to certain patents? That would be a massive fuck up if AMD allowed that. I could understand an agreement where they’re both able to use certain patents.

    • data8504
    • 7 years ago

    Jeez. 40b PA (physical address)? That’s only 1 TB of addressable space… (DRAM, MMIO, etc.) Pretty skimpy for a “64-bit” processor. Even the 44b big brother is way small for the enterprise. Maybe I’m missing some trick they’re pulling…

      • just brew it!
      • 7 years ago

      The “trick” is that your server will have dozens or hundreds of these things in it. It’s not designed for workloads that require a small number of big beefy nodes.

      • chuckula
      • 7 years ago

      1. 99% of these chips will end up in your smartphone/tablet, and while smartphones will certainly go beyond 4GB of RAM in the next few years, I’m not holding my breath waiting for them to hit 1TB of RAM.

      2. People hear the word “ARM” in association with a server and think that ARM is attacking your 8-way Xeon mega-server. Not in the slightest. The ARM server chips are stacked onto small blades with relatively small amounts of RAM that are in a shared-nothing architecture. Think of what SeaMicro is selling right now: A huge rack full of Atom blades that all share a common I/O backend but are otherwise independent servers with their own local pool of RAM. Just substitute an ARM chip for the Atom and that is exactly what an ARM server solution will look like. No way each blade is going to approach 16 TB (or even 1TB) of RAM within the useful lifespan of these chips.

      • kalelovil
      • 7 years ago

      A 44bit physical address space is hardly “way small for the enterprise”.
      It is the same as that used by the Itanium 2 and Nehalem-EX, for example, and allows up to 16TB of directly addressed memory for each coherent system.

      Supercomputers and large datacentres which may need more memory than this are comprised of individual nodes and a distributed memory architecture and thus avoid the directly addressable memory limitation.

    • maxxcool
    • 7 years ago

    Hate and -1 all you want, This is another nail, not step forward. More a step sideways to survive…

      • willmore
      • 7 years ago

      What’s that nail in?

        • derFunkenstein
        • 7 years ago

        His forehead, because I’m not sure he’s sure exactly what he means, let alone whether or not it’s grounded in reality.

      • maxxcool
      • 7 years ago

      Coffin…

      • rrr
      • 7 years ago

      You nailed it. And I nailed your gf.

      Who’s better off?

    • tootercomputer
    • 7 years ago

    So what role does AMD play exactly? Where do they fit in?

      • just brew it!
      • 7 years ago

      High speed interconnects. (They know how to do them, they invented HyperTransport.) Their plan is to use ARM as the basis for massive racks of inexpensive blade servers.

      Remember their acquisition of [url=http://www.seamicro.com/<]Seamicro[/url<] a little while back? They've been planning this.

        • lilbuddhaman
        • 7 years ago

        But they weren’t originally planning on it being the only egg left in their basket.

          • just brew it!
          • 7 years ago

          Who says they’re planning that now?

            • MadManOriginal
            • 7 years ago

            [quote<]Who says they're planning now?[/quote<] FTFY!

            • just brew it!
            • 7 years ago

            My previous post is still valid even in that context…

            • derFunkenstein
            • 7 years ago

            I think you have to intentionally misread the press release Monday to think AMD is going ARM-only.

    • chuckula
    • 7 years ago

    [quote<]A month later, GloFo announced that its 14-nm FinFET process will be ready in 2013. Perhaps the first processors based on the Cortex-A50 series will use that process.[/quote<] Sorry, that ain't gonna happen. Any Cortex A5x chips that make it to market in 2014 will be on a 20nm process.

      • mnecaise
      • 7 years ago

      If I recall correctly, ARM’s slides said Cortex A50 will be released in 20nm FinFET first. So, you’re probably correct.

      edit: oops slides show core IP availability for 28nm and 20nm, with 20nm coming in 2013 and 14nm later.

    • Arclight
    • 7 years ago

    [quote<]According to ARM, the Cortex-A57 will deliver the "highest single-thread performance," and systems featuring it will have performance "comparable to a legacy PC."[/quote<] Wonder what that means....pay more to get less?

      • ronch
      • 7 years ago

      And we thought Bulldozer was a step back.

        • flip-mode
        • 7 years ago

        Aren’t legacy PCs (Conroe, Penryn, Nahalem, Deneb, Thuban) faster than Bulldozer?

      • chuckula
      • 7 years ago

      Let’s use the marketing translation machine on this one:
      “Performance comparable to a legacy PC” –> We’re faster than a Pentium IV now!

        • jjj
        • 7 years ago

        You guys are way too negative.
        A57 should be 15-30% faster than A15 on same process and running 32bit code.Go 20nm and native code and you get quite a bit more. Don’t forget that this is a core that goes in phones and A15 seems to be a huge step forward (anandtech published some perf numbers for A15 in their ChromeBook review today).
        A53 is touted as just over half a mm2 size (no idea on what process) while being faster than A9 so it’s worth noticing too.

        “Perhaps the first processors based on the Cortex-A50 series will use that process.”
        FinFET comes after 20nm,nobody will be ready with it before so there is about 0 hope for that,If you are talking first AMD SoC,not in general first A57 SoC then it’s unlikely only because they are talking 2014 not 2015.

          • chuckula
          • 7 years ago

          From what we’ve seen of actual A15 silicon, if all of these things pan out then the Cortex A57 in 64 bit mode will have an IPC that is comparable to a Core 2 with a few exceptions*. That’s actually very very good and if ARM can keep the power envelope reasonable, then they will have quite a bit of success in tablets & phones. What it won’t do is take a big chunk out of the mobile PC market or the server market. It also won’t prevent Intel from competing in tablets & phones with 14nm Atom chips that will also have a radically different architecture than the 2008-era designs that are on the market now.

          * Exceptions: The 128 bit NEON has been improved in the new ARM setup but is still not as good as SSE (even the versions of SSE in the Core 2). In super vectorized code the IPC of the Core 2 is still ahead of the 64 bit ARM, but in most other operations they will be comparable.

          • Sam125
          • 7 years ago

          Judging from ARM’s slides it looks like the A57 looks like it’s going to be about 80% faster than an equivalent A15 while the A53 is going to use about 10-25% less power than the A15.

          In a big.LITTLE configuration, ARM might find that it has a huge winner in ultra-dense micro servers.

          At the very least, ARMv8 (Cortex-A50) is a good sign that smartphones are only going to continue to improve as well. =)

        • cygnus1
        • 7 years ago

        i think when they say legacy PC, they mean you’re getting the performance of something that used a couple orders of magnitude more power than the product they’re delivering with their, for them, normal low power envelope

          • Sam125
          • 7 years ago

          Are you sure it’s not a cheeky statement to mean that “legacy PC” being an x86 based PCs? =P

      • Alexko
      • 7 years ago

      Why “pay more”?

      ARM chips aren’t usually expensive.

      • just brew it!
      • 7 years ago

      More like “pay less to get something that can be used to build an inexpensive, passively cooled, I/O-centric blade server module”.

        • Arclight
        • 7 years ago

        Can you be certain that that will be the case in 2014? And why should AMD be the one offering such a product at the best price/performance when we know there will be others licensees?

          • just brew it!
          • 7 years ago

          No, of course not. In fact, as I’ve indicated in comments on the other AMD/ARM stories I think it is a risky bet. I was just trying to explain what their strategy appears to be.

      • Deanjo
      • 7 years ago

      Hmmm, my Tandy 1000 EX is a legacy PC…..

      • Flatland_Spider
      • 7 years ago

      It means the the A57 has better single-thread performance then the A53, and other ARM processors, and x86 is old and busted while ARM is the new hotness.

      I wouldn’t give any credence to that being a statement about performance. It’s a product launch, and they’re hyping up their product.

Pin It on Pinterest

Share This