New Atom chips headed to servers

Intel’s Atom processor lineup made a name for itself in netbooks. Since then, it has since branched out into tablets and even smartphones. Now, Intel has released a new family of Atom processors, the S1200 series, aimed squarely at servers. These babies have 64-bit addressing, ECC memory support, and other perks that make them suitable for heavy-duty applications—and they fit inside power envelopes as small as 6.1W.

Here’s Intel’s pitch for the new SoCs:

The SoC includes two physical cores and a total of four threads enabled with Intel® Hyper-Threading Technology2 (Intel® HT). The SoC also includes 64-bit support, a memory controller supporting up to 8GB of DDR3 memory, Intel® Virtualization Technologies (Intel® VT), eight lanes of PCI Express 2.0, Error-Correcting Code (ECC) support for higher reliability, and other I/O interfaces integrated from Intel chipsets. The new product family will consist of three processors with frequency ranging from 1.6GHz to 2.0GHz.

Atom S1200-series processors are shipping today. Intel’s price list quotes asking prices ranging from $54 for the S1220, which runs at 1.6GHz with a 8.1W TDP, to $120 for the flagship S1289, which is clocked at 2GHz and has a 14W TDP. (The lone 6.1W offering has a 1.6GHz clock speed and costs $64.)

Intel says it already has "more than 20" design wins, which include micro server, storage, and networking systems. Among the early adopters are Accusys, CETC, Dell, HP, Huawei, Inspur, Microsan, Qsan, Quanta, Supermicro, and Wiwynn.

The Atom S1200 family is based on current-generation silicon and fabbed on a 32-nm process. Next year, however, Intel plans to deliver new, 22-nm Atom server chips code-named Avoton. Customers can look forward to "world-class power consumption and performance levels," the company claims. As we reported yesterday, Intel has developed a special version of its 22-nm process for system-on-a-chip devices like the upcoming Atoms. Consumer "Silvermont" Atom processors will also be manufactured using that process.

Comments closed
    • chuckula
    • 7 years ago

    Too little too late… the world has moved on!

    Thank you mostly-inaccurate, we needed that insight. I’m just waiting for their $50 report on how ARM has invented wormhole technology that allows their 64-bit chips that might be on sale in 2014 to be brought back in time to 2012 in sufficient quantities to completely dominate the server market and drive Intel out of business.

    Because we all know that a theoretically available chip in 2014 is going to totally outperform some vaporware so-called x86 chip that is so-called actually available in so-called servers in the so-called year of 2012… please Intel, what kind of sheeple do you take us for?

    • NeelyCam
    • 7 years ago

    Looks like AMD got pissed off:

    [url<]http://www.tomshardware.com/news/amd-intel-atom-s-arm-microserver,19701.html[/url<] Looks like they even got their proxy, SemiAccurate, to do their bidding... AMD: [quote<]"Intel is finally acknowledging the micro server party, although we do wonder if itโ€™s 'too little, too late?'"[/quote<] S|A: [quote<]"Too little, too late, the world has moved on"[/quote<] AMD: [quote<]"Intel is [b<]way[/b<] behind on small cores. They have no cell phone market share, little tablet market share, and now they are threatened that they will lose server market share. AMD and its SeaMicro technology are leading the charge in micro server technology and development."[/quote<] S|A: [quote<]"Intelโ€™s myopically missed this nascent market in a bid to protect big chip margins, and will pay the price in a few quarters. The world moved around Intel, and now no one cares."[/quote<] [url<]http://semiaccurate.com/2012/12/11/intel-announces-centerton-the-server-atom/[/url<] Most interestingly, both S|A and AMD are suggesting AMD is well ahead Intel in this space, yet they have [i<]nothing to offer[/i<] themselves at the moment. If anything, it looks like Intel is well ahead

      • chuckula
      • 7 years ago

      Hey! AMD is [b<]way[/b<] ahead of Intel in this space! They offer SeaMicro servers with freedom-fries-fabric! Name one server that you can buy directly from Intel that has the same features... YOU CAN'T!! [sneaks back to PR department to shred all of the documents for SeaMicro showing that all of those servers are already running Atoms that weren't even designed with the low-power server market in mind]

        • NeelyCam
        • 7 years ago

        Yeah.. Didn’t SeaMicro try to get bought by Intel? Intel said “no; we have something better in-house”, so they had to go for AMD instead..

        Overall, they sound a bit like a jilted ex.

        “No – let me get some facts straight: [b<][i<]I[/i<][/b<] broke up with [b<][i<]YOU[/i<][/b<]! And Freedom Fabric is the best name [i<]evah[/i<]!! Besides, you suck in bed!! There! *sniff* *sniff* ..please come back... I miss you... *sadface* "

      • MadManOriginal
      • 7 years ago

      [quote<]They have no cell phone market share, little tablet market share, and now they are threatened that they will lose server market share.[/quote<] That sentence is hilarious because what I get from it is two things: 1) Intel has huge untapped and quickly growing markets into which they will grow 2) Intel's server market share is so dominant that [i<]any[/i<] alternative in the server market means Intel loses market share.

        • bcronce
        • 7 years ago

        “Intel’s server market share is so dominant that any alternative in the server market means Intel loses market share.”

        The opposite of “once you hit rock bottom, the only way to go is up”. Once you hit the top, the only way to go is down.

          • NeelyCam
          • 7 years ago

          [quote<]The opposite of "once you hit rock bottom, the only way to go is up". Once you hit the top, the only way to go is down.[/quote<] In AMD's case, once you hit rock bottom, you can keep digging to get even deeper. In Intel's case, once you hit the top, you can start building a tower to reach even higher

      • ludi
      • 7 years ago

      IOW the spirit of Jerry Sanders III lives on: if you’re going to crash and burn, flap your yap the whole way down.

      • d0g_p00p
      • 7 years ago

      That’s not true at all. I have a HP MicroServer doing FreeNAS duties with a AMD Turion CPU in it. The previous gen used a NEO CPU. AMD has been doing the microserver deal for awhile now.

        • MadManOriginal
        • 7 years ago

        That’s not the kind of microserver this is really targeting.

    • NeelyCam
    • 7 years ago

    Intel just took over the only server segment where ARM had a chance.

      • brute
      • 7 years ago

      Confucius say: Server without ARM not much of server at all. Should be hostess.

      • MadManOriginal
      • 7 years ago

      Almost, but not yet…these are still old-fashioned Atom cores which seem to be slower than the ARM A15-class chips with the same thermals (see reviews of the ARM Chromebook.) However, Silvermont with OoO and then 14nm will certainly shackle ARM’s chances. Since it takes a while for designs to become reality and the saying that ‘no one ever got fired for buying Intel’ there is a good chance ARM will never get a big foothold in the server space.

        • NeelyCam
        • 7 years ago

        Actually, the ARM Chromebook review shows that ages-old 45nm Atom [i<]CPU[/i<] is more efficient than the A15 28nm part. A15 has better performance, but much higher power consumption under load The old Atom platform consumed a lot of power, presumably because the chipset wasn't really optimized for Atom systems.. that's why the numbers in the review are high for the Atom system. But if you look at the [i<]differences[/i<] between idle and load power numbers, you'll see a very different story. 32nm Atom probably spanks 28nm A15 in efficiency. At 22nm it's not even a contest. It's very difficult to scale performance up without losing efficiency (while it's easy to gain efficiency when scaling performance down). ARM is unable to hold on to it's high efficiency halo when they are trying to move to PC and server performance space. I pointed this out when Anand wrote an article about big.Little: [url<]http://www.anandtech.com/show/4991/arms-cortex-a7-bringing-cheaper-dualcore-more-power-efficient-highend-devices/2[/url<] See the second figure? "Rich Web Services" in the A15 column shows a 10% efficiency improvement. The problem is, this is comparing 40nm A9 to 32/28nm A15. The process shrink alone should have improved efficiency more than that (i.e., a 32/28nm A9 would be more efficient than the A15). The A15 architecture just looks less efficient than A9... more powerful, sure, but less efficient. And perf/watt is king in this microserver space Meanwhile, Intel has a much easier situation: take 40nm Atom, shrink, drop supply voltage, integrate chipset, do small tweaks, and voila - you have a much more efficient system.

    • brute
    • 7 years ago

    they’re sending the chips to servers, but no mention of dip?

    i will not tip. dry chips are the bane of my restaurant experience

      • NeelyCam
      • 7 years ago

      Use ketchup

    • JMccovery
    • 7 years ago

    Has anyone found what makes this Atom a ‘SoC’? Even Ark doesn’t have detailed specifications of the S1200 Atoms.

    • dpaus
    • 7 years ago

    I wonder how many of those design wins are using AMD’s Seamicro architecture? ๐Ÿ™‚

    • Chrispy_
    • 7 years ago

    It’s a good start, but it’s still the underperforming, inefficient and undesirable Bonnell architecture.

    We’re waiting for Silvermont and Atom will continue to be a relatively bad product until that new architecture arrives.

      • willmore
      • 7 years ago

      In which case it might be performance/power competetive, but it’ll still be more expensive than anything else.

        • NeelyCam
        • 7 years ago

        [quote<]it'll still be more expensive than anything else.[/quote<] Price goes hand-in-hand with quality metrics (in this case performance/power). If it's merely "competitive", it'll be priced as such.

          • willmore
          • 7 years ago

          Then explain these chips. They’re well more expensive than a comparable ARM part and have poorer performance/watt.

          Take a look at the performance numbers over at phoronix.
          [url<]http://www.phoronix.com/scan.php?page=article&item=calxeda_ecx1000_atom&num=1[/url<]

            • NeelyCam
            • 7 years ago

            [quote<]have poorer performance/watt.[/quote<] Um... the article says: [quote<]"However, for this article there are unfortunately not any performance-per-Watt results to share."[/quote<] They keep listing the supposedly measured numbers for a few Calxeda cores (not the whole system, mind you), and repeating that D525 has a TDP of 13W. Sort of reminds me of this: [url<]http://www.theregister.co.uk/2012/08/13/xeon_vs_calxeda_arm_apache_bench/[/url<] So, without [i<]measuring[/i<] the power consumption, those Phoronix numbers don't mean much.

    • hasseb64
    • 7 years ago

    Wonder how much IDLE consumtion this SoC will have.
    My Sandy Bridge 4 core idle at 4W today (GPUand CPU), my guess would be around 7-8 W for my “SoC” total.
    See ECC support as a new nice feuture!

      • Sahrin
      • 7 years ago

      If your server is idling then you are wasting capex.

        • Celess
        • 7 years ago

        Right on!

        • brute
        • 7 years ago

        “You are not to make for yourselves an idol, or any likeness of what is in heaven above, or on earth below, or in the water under the earth. ”

        it is also a sin

        • bcronce
        • 7 years ago

        If your system is at 80% load, it is 20% idle. Not everyone has a fully clustered setup. The cost to design and manage a properly implemented VM setup is typically more expensive than letting a few servers sit idle for portions of the day.

    • sircharles32
    • 7 years ago

    I wonder if these will be powerful enough to handle software based RAID 5. After all, these are going into low power servers, where an added hardware based controller would eat up an additional 10-20 watts, which could be significant.

      • DancinJack
      • 7 years ago

      Considering they are powerful enough, albeit with dedicated hardware, to do some virtualization, I can imagine they will do RAID 5 just fine. Not that anyone in a data center would want to do software RAID 5 though.

      • bcronce
      • 7 years ago

      RAID5 is old and it is recommended to use RAID6 or higher(2 or more parity drives).

      Personally, I would not use an ATOM cpu on a file server if performance is a concern. I would mostly use them for Web servers, DHCP/DNS/etc, routers, IDS, NOSQL shards, pretty much anything that is network heavy and CPU/memory light.

        • bcronce
        • 7 years ago

        stdRaichu said the same thing and got +1 and I got -1. I love TR’s grief system.

          • NeelyCam
          • 7 years ago

          There is a subtle difference. stdRaichu said he used RAID6 without any commentary on RAID5. You implied RAID5 is a bad idea (indirectly challenging sircharles32’s plans) and that people should use RAID6 instead. You didn’t even explain why – you just sort of stated that as your expert opinion.

          So, as far as I can tell, that was the reason why you got -1 and stdRaichu did not.

          • stdRaichu
          • 7 years ago

          FWIW I agree with you about RAID5, but as Neely points out you’ll need to qualify it ๐Ÿ˜‰ So for everyone else out there…

          Obviously the big problem with any RAID is rebuild time, and since RAID5 can only tolerate a single failure, if another drive fails during a rebuild, you’re toast. For RAID5 arrays with only a few spindles (my general rule of thumb is that I’ll only use RAID5 on an array with three or four drives), the chances of this happening are slim, but as with RAID0, each additional disc you add massively increases your chances of failure, and RAID6 is the only way of mitigating that. The backup array is eight 2TB drives of varying quality and manufacturers and I have had a dual failure during a rebuild, so RAID6 saved my arse then.

          Naturally, RAID10 is still the best if you want performance (especially random IO!) as well as resilience and scalability (including resilience that scales with the number of spindles), but seemingly no-one other than me ever uses it since it’s quite expensive in terms of spindles.

      • stdRaichu
      • 7 years ago

      I’ve got an atom D525 in my QNAP, and I found it painfully slow for any parity RAIDs. I eventually switched it to RAID10 instead simply to get rid of the CPU overhead.

      Now that I’ve found a suitably small hot-swap case, I’m rebuilding my backup server (with eyes on replacing my NAS with a very similar setup), with the i5 3470T it’ll easily break 300MB/s seq r/w in RAID6 and uses only slightly more power than the atoms (and the same amount when idling). ECC is available in the same power envelope and form factor, just costing slightly more spondoolix.

      Oodles more power and still costs less than a QNAP ๐Ÿ™‚

    • Shouefref
    • 7 years ago

    Am I wrong, or could this lead to interesting energy savings?

      • MadManOriginal
      • 7 years ago

      Both.

      • hoboGeek
      • 7 years ago

      Yes.

Pin It on Pinterest

Share This