Report: Intel cooking up eight-core Atom for servers

ARM-based CPUs are working their way into servers, and Intel may answer with an eight-core Atom designed for the same space. So says this story over at SemiAccurate, which claims the chip will be known as Avadon or Abadon. The CPU will reportedly be based on a Silvermont Atom core that “people who know” describe as a “mildly out of order” design. This new core apparently offers 20-25% better clock-for-clock performance than the existing Atom.

Silvermont is slated to be built using 22-nm process technologyโ€”tri-gate transistors and all. SemiAccurate says the eight-core chips won’t arrive until late 2013, though. That seems like a reasonable timeline given the fact that the first Silvermont-based Atoms, which will have fewer than eight cores and be targeted at netbooks and tablets, are expected in 2013. While Intel didn’t mention an eight-core version at IDF a couple of months ago, it did confirm that Silvermont would debut in 2013.

As SemiAccurate points out, an eight-core Atom arriving two years from now will probably face plenty of competition from the ARM camp. Server-oriented CPUs based on Nvidia’s Project Denver effort may be available by then, as well. Intel will have x86 compatibility on its side, of course, but that may become less important as the popularity of the ARM architecture grows.

Comments closed
    • StuffMaster
    • 8 years ago

    What about eight-atom cores? When can we have those?

    • Austin
    • 8 years ago

    ;o) Oooo … 8 core Atom … that should be good enough to run something like a kettle or a toaster. Yay!

      • dpaus
      • 8 years ago

      A 2-slice or a 4-slice toaster?

        • Yeats
        • 8 years ago

        4 slice toaster, but you can fit 2 smaller slices of bread in each slot.

        Oh wait, that’s the Bulldozer toaster, my bad…

      • ybf
      • 8 years ago

      A kettle needs to sample the temperature and switch the current.

      8 2-GHz cores could do sample and switch about 8 billion times a second.

      One could run enough kettles for every human to have a hot can of whoopass for lunch.

        • bhtooefr
        • 8 years ago

        Yeah, but it’s an Atom, so it only has high enough IPC to actually run the sample and switch routine every few minutes. ๐Ÿ˜‰

        (OK, OK, that’s exaggerating, but…)

        • Farting Bob
        • 8 years ago

        Yea, but what about people with server racks full of kettles?

          • dpaus
          • 8 years ago

          This would fit them to a ‘tea’…

            • UberGerbil
            • 8 years ago

            Unfortunately the learning curve is steep.

            • bitcat70
            • 8 years ago

            This is such a steamy thread…

            • pedro
            • 8 years ago

            Comment of the week. Tips hat.

    • chuckula
    • 8 years ago

    What I love about SemiAccurate is how they paint this picture as Intel desperately trying to gain a foothold in the server world against an unstoppable ARM force. Now the only thing left is for ARM servers to actually show up, which, even according to the latest press releases from HP, won’t be happening for some time and it is very unclear how much traction they’ll gain when you realize that the power consumption of the server is often dictated by much more than just the CPU. Last time I checked, RAM, networking, disk controllers, etc. didn’t magically consume lest power because they were hooked up to an ARM chip.

      • flip-mode
      • 8 years ago

      Yeah it’s certainly an odd way of presenting it.

      • Chrispy_
      • 8 years ago

      This.

      Based on what I know is in my server racks, and what my UPS says each rack is drawing, I would guess that about a third of a typical server is CPU draw.

      Those things have fans, disks, Ram, RAID controllers and various other pieces of silicon consuming power too.

        • OneArmedScissor
        • 8 years ago

        [quote<]Those things have fans, disks, Ram, RAID controllers and various other pieces of silicon consuming power too.[/quote<] Much of which would be eliminated with a SoC-based server, which is the entire point...

          • LoneWolf15
          • 8 years ago

          They put fans, disks, and RAID controllers in the chips now?

            • NeelyCam
            • 8 years ago

            much != all

            • BobbinThreadbare
            • 8 years ago

            Remove fans, disks, RAM, and Raid controllers from fans, disks, RAM and Raid controllers and you are left with what?

    • nafhan
    • 8 years ago

    So… Ivy Bridge is going to be 15 to 35w. What kind of power envelope will this have? Very interesting.

    • dashbarron
    • 8 years ago

    It’s kind of ‘funny,’ but makes sense they’d push more cores and multi-core applications on mobile devices than cranking up higher and consequently faster and hotter/more energy-draining processor. Obvious observation.

    I’m sidelining into: I really hope this pushes for more rapid development of parallelism and multi-cored applications, then Intel can let go of the floodgate and push out their 12 core desktop processor to the public.

      • adisor19
      • 8 years ago

      This is all about visualization. More cores = more VMs.

      Ado

        • MadManOriginal
        • 8 years ago

        Thanks for the clarification, silly old me always thought VM stood for Virtual Machine!

        • BobbinThreadbare
        • 8 years ago

        Why do you need more cores to run more VM when Intel and AMD have all those VT extensions?

          • UberGerbil
          • 8 years ago

          Why do you need more than one core to do computations when Intel and AMD have all those SIMD extensions?

      • UberGerbil
      • 8 years ago

      “Rapid development of parallelism and multi-cored applications” is not waiting on, nor will it be affected by, the existence of 8 core Atom server chips. The applications that can take advantage of this — server apps and virtualization — are already well-threaded. The other applications –the ones you probably care about — may very well be about as threaded as they can be already also. Developers in 2011 for the most part are not sitting around with lots of potential thread-level parallelism in their code that they’ve chosen not to exploit; even if they were, the availability of [i<]octo-core Atoms[/i<] is not going to suddenly motivate them to thread their code. In many cases, there's no latent TLP to exploit, or what little there is isn't worth the headaches. Lots of code ends up serialized waiting on some shared resource, and lots of code isn't CPU-bound anyway. If an app only uses a couple % of the CPU, is it really worth re-coding it just so it uses a single % across each of several cores? And if you did, you'd probably find that the net throughput isn't any higher, because the threads are spending a lot of their time waiting on each other. Amdahl's Law is a harsh mistress. Intel is certainly planning 12+ core chips, but they're not headed to the desktop -- and that's not because some lazy software developers haven't got around to threading their desktop apps.

    • Arclight
    • 8 years ago

    I wonder how FX 8150 fares against this CPU….would be hilarious if the atom came close to the AMD chip.

      • tfp
      • 8 years ago

      Not fair 16 threads vs 8 ๐Ÿ˜‰

        • NeelyCam
        • 8 years ago

        Eight real cores against eight “real” cores. Fair is fair.

      • kamikaziechameleon
      • 8 years ago

      That it would. I expect the single core performance per clock to be near identical. ๐Ÿ™

        • OneArmedScissor
        • 8 years ago

        It says 20-25% better per clock right there in the article. That’s barely even existing Bobcat territory.

        There’s no reason to expect miracles of the 22nm Atom. Haswell comes out in SoC form first.

        • ImSpartacus
        • 8 years ago

        You’re probably correct.

        We’re talking about a two year jump (this is a late 2013 chip). It’s like comparing a typical ULV Sandy Bridge (to the highest binned 2009 Penryn chip (the 3.07GHz 45W T9900). The SNB part uses at least half (probably a third) of the power, but will at least match the Penryn chip in performance.

        I know the power deficit is much bigger with next-gen Atom and 2-billion-transistor Bulldozer, but Bulldozer has [i<]really[/i<] bad single threaded performance.

      • EV42TMAN
      • 8 years ago

      $20 says the atom beats it

        • Arclight
        • 8 years ago

        If you lend me $20 i’ll do it for the lulz ๐Ÿ˜€

          • swaaye
          • 8 years ago

          It would have been gut busting to see this 8 core Atom launch say a week after Blowdozer and match it in benches. ๐Ÿ˜‰

    • NeelyCam
    • 8 years ago

    “offers offers” -> “offers”

      • dpaus
      • 8 years ago

      No, they mean that it really, really offer offers better better performance.

        • Deanjo
        • 8 years ago

        No, no, no, they mean “much more betterest performance”. ;P

      • ybf
      • 8 years ago

      In the same way that “pizza pizza” is “pizza”. I.e., saying it twice doesn’t necessarily make it true.

Pin It on Pinterest

Share This