Forgot your password?
typodupeerror
Hardware

SeaMicro Unveils 512 Atom-Based Server 183

Posted by CmdrTaco
from the why-stop-there dept.
1sockchuck writes "Stealthy startup SeaMicro has unveiled its new low-power server, which incorporates 512 Intel Atom CPUs, a load balancer and interconnection fabric into a 10u server. SeaMicro, which received a $9.3 million government grant from DOE to develop its technology, says its server uses less than 2 kilowatts of energy — suggesting that a single rack with four SeaMicro units and 2,048 CPUs could draw just 8 kilowatts of power. Check out the technical overview, plus additional coverage from Wired, GigaOm and VentureBeat."
This discussion has been archived. No new comments can be posted.

SeaMicro Unveils 512 Atom-Based Server

Comments Filter:
  • by steelcobra (1042808) on Monday June 14, 2010 @11:03AM (#32565296)

    Wow. Just think how many servers you could run in VMWare on that. A hundred would be a decent functional number.

    • You are talking about the Atom here. It is sooooo slowwwwww!
      • Re: (Score:2, Insightful)

        by sleeping143 (1523137)
        I think the more important element here is the performance/energy consumption ratio. Atoms might be slow, but they're not so slow that their minuscule power consumption can't make up for it.
        • Re: (Score:3, Informative)

          by afidel (530433)
          Which would you rather have in 8kw, 2048 atom cores or 768 56xx cores? Because that's what HP can cram into 4xc7000's in the same power and floor space envelope (not sure about capital outlay though). Personally for my workloads I'd take the 56xx cores, but YMMV.
    • The ATOM doesn't support virtualisation in hardware, so at best, you are limited to one 32bit OS as a guest per core (no 64bit since you need CPU support for virtualisation) - and even then, the performance is so bad that you might as well just not bother. Having tried virtualbox on an Atom330, I can assure you it's really not worth waiting for the guest OS to finish installing....
    • Re:Vitual center (Score:4, Interesting)

      by AigariusDebian (721386) <aigarius@debLISPian.org minus language> on Monday June 14, 2010 @12:00PM (#32566060) Homepage

      Actually it is much more interesting to handle each of them as you would handle an individual virtual machine - so you have 512 nice low-powered virtual servers with each of them having a fixed and dedicated processor.

      In fact such a load-out would be very useful for hosting companies - you can have a ton of small clients with minimal management or scheduling burden.

  • by TeXMaster (593524) on Monday June 14, 2010 @11:07AM (#32565354)
    The question is, how good is the performance for, say, intensive numerical computations? Is the gigaflop per watt convenient?
    • by Nadaka (224565) on Monday June 14, 2010 @11:17AM (#32565486)

      The FLOPS are not all that important for this device. It isn't designed to crunch big numbers. It is designed as a web|web application server with the goal of serving far more connections per watt than a traditional server.

      • by jandrese (485) <kensama@vt.edu> on Monday June 14, 2010 @11:27AM (#32565618) Homepage Journal
        But traditional web servers aren't CPU bound, they're IO bound at high connection rates. It might help if you need to do a whole lot of https traffic, but even then this smells of overkill. If you're really planning to use this as a webserver, I'd be a whole lot more interested in the IO backplane and the available IO ports to the server.
        • System Specs (Score:5, Informative)

          by dlapine (131282) <(ude.cuiu.ascn) (ta) (enipald)> on Monday June 14, 2010 @11:53AM (#32565954) Homepage

          This is a good start- SM10000 System Overview [seamicro.com]

          Interconnect is 1.28 Tbps or 2.5 Gbps per core.

          I/O includes a minimum or 8 gige or 2 10-gige, which can be increased to 64 gige or 16 10-gige links per chassis.

          This unit runs as 512 system images using stock 32 bit OS's. Each CPU may have 1 or 2 GB's of ram and up to 64 local drives may be installed and divided among the CPU's with the included management software. The unit supports PXE boot, so the system images may run off local disk or from a ram image.

          Just to note, the Atom z530 is a single core, 32 bit only CPU, if that matters.

          I couldn't tell you if the 16 10-gige links would seriously limit this box or not. You'd have to show me a data center with more than 160 Gbps of internet connectivity first. :) And that's assuming you only purchased one of these suckers, because you'd need that much per chassis.

          • Yikes! Front access for the hard drives, rear access for the power and network connections, and side access for the hot-swappable CPU boards. Maybe there's a way to mount that beast of a chassis on full-extension slides to get side access, but that'd make the rear wiring a nightmare. Requiring an access path on the side of each rack would reduce it's footprint efficiency by half. Doesn't quite fit into your typical data center floorplan.
            • by dlapine (131282)

              Hmmm, didn't see that. Given that it takes the place of two full racks, maybe you're supposed to put it on a pedestal in place of them.

              Something like this [exhibitsusa.com] for easy access.

              Or, maybe you could rotate it 90 degrees and mount it CPU-access-side up. At 10U that's only 15", so it should fit in a 19" rack. :-)

              Seriously, if you don't plan to do hot swap on the CPU boards, you'd be OK in a normal rack. I'm not sure I'd trust hot swap for CPU boards anyways.

      • by imgod2u (812837)

        If that's the case, Atom is a horrible choice for the CPU. Atom's strongest feature is its 128-bit vector (SSE) unit for SIMD FLOPS.

        The in-order (albeit SMT-enabled), 2-issue integer/LDST/Conditional pipeline is lackluster at best from either a performance or a performance/watt standpoint.

        • If that's the case, Atom is a horrible choice for the CPU. Atom's strongest feature is its 128-bit vector (SSE) unit for SIMD FLOPS.

          But it's a poor choice for high-precision scientific calculations. The vector unit on the Atom is optimized for 32-bit Floats, and when you switch-up to Doubles, the performance drops off the map.

          See this performance comparison (Processor Multimedia) [pcper.com]. The dual-core Atom 1.6 is actually faster than the dual-core Athlon 64 1.6 when running SSE 32-bit Floats (this is due to the

          • by imgod2u (812837)

            But it's a poor choice for high-precision scientific calculations. The vector unit on the Atom is optimized for 32-bit Floats, and when you switch-up to Doubles, the performance drops off the map.

            See this performance comparison (Processor Multimedia). The dual-core Atom 1.6 is actually faster than the dual-core Athlon 64 1.6 when running SSE 32-bit Floats (this is due to the dual 64-bit SSE units on the Athlon 64, and poor scheduling on AMD's part). But when you transition to Double precision, the performance on the Athlon 64 only drops by roughly 2x, whereas the Atom sees a 5-6x performance drop!

            This means you have very little flexibility in workloads you can use - if you're not using 32-bit precision, you've wasted your money.

            That was for x87 FP though. While it's important in that a lot of code out there might still use it, for performance on any Intel architecture, you're supposed to use SSE (scalar SSE if you're not doing SIMD). I would like to see SSE performance broken down to SP vs DP. I suspect, from what little I know of the vector pipeline, that it'd be pretty much exactly half the performance.

            It's not so bad if you're running a highly-multithreaded program on the system. The Atom uses the SMT to great advantage, in most cases seeing a %30 performance increase. But yeah, for single-threaded code, the Atom is a joke.

            Even for highly threaded programs, the performance/watt will not be competitive with (lightly) OoOE architectures. In-order is s

            • That was for x87 FP though. While it's important in that a lot of code out there might still use it, for performance on any Intel architecture, you're supposed to use SSE (scalar SSE if you're not doing SIMD). I would like to see SSE performance broken down to SP vs DP. I suspect, from what little I know of the vector pipeline, that it'd be pretty much exactly half the performance.

              No. The Multimedia benchmark in the Sandra suite is all SSE. See their FAQ here [sisoftware.net].

              The following is their only Double test:

              Q: Wha

      • by drinkypoo (153816) <martin.espinoza@gmail.com> on Monday June 14, 2010 @02:03PM (#32567904) Homepage Journal

        Then they screwed up, and they should have used ARMs, because a great deal of Atom's performance lies in its multimedia instruction set. Or in other words, if you're not pushing flops, you have a lot of hardware lying around unused. Atom delivers a lot of flops (or iops, for that matter) but doesn't shovel data any more efficiently than anyone else.

    • by deadline (14171)
      Here are some numbers from a post [slashdot.org] I made last week.
    • by BitZtream (692029)

      Why exactly would you be doing real processing on a virtual machine?

      If you need real processing power you don't put it on a VM, thats just dumb and wasteful.

      Virtual machines are for testing and silly little one off instances of something that some department 'needed' on a 'server' that gets used by 3 people, twice a year, and thats only until next month when they've forgot about it completely.

      • by drsmithy (35869)

        Why exactly would you be doing real processing on a virtual machine?

        Because VMs are much, much nicer to manage than real hardware, if for no other reason than making the ideal one-service-per-server scenario less wasteful.

        If you need real processing power you don't put it on a VM, thats just dumb and wasteful.

        The overhead of modern virtualisation software on modern hardware is negligible (a couple of percent, if that, outside of corner-case workloads).

        Virtual machines are for testing and silly little

  • by mark-t (151149) <[markt] [at] [lynx.bc.ca]> on Monday June 14, 2010 @11:09AM (#32565378) Journal
    ...misread the headline as saying that somebody had made a server out of only 512 atoms (as in the particle, not the cpu)?
    • I am with you on that one, and for a second I thought that we were going to be reading about some kind of technological revolution.
    • I want my server out of 512 actual atoms! *waaaahhhh*

      P.S.: I bet a ARM-based server would beat the crap out of that (TFA) thing on the price/performance and price/energy scales

    • When I first read the headline I thought "Wow--someone's just commercialized quantum computing!"

    • by Opyros (1153335)
      And 512 atoms ought to be enough for anyone—
  • Imagine how may of these servers you could fit in the space of even a single grain of rice, let alone a standard 1U enclosure!

  • by Chas (5144) on Monday June 14, 2010 @11:12AM (#32565426) Homepage Journal

    Oh, wait. Nevermind.

    • Re: (Score:3, Insightful)

      by CAIMLAS (41445)

      It would seem that the 'beowulf cluster' is starting to fall out of style, doesn't it? :P We're getting to the point where such concepts are as quaint as a "Cray supercomputer" were just a couple years ago.

      • by TheRaven64 (641858) on Monday June 14, 2010 @11:25AM (#32565592) Journal
        Clusters are still very much alive. They're cheap to build and give you a lot of computing power to play with. If anyone mentions Beowulf when describing them, however, it's a good clue that they have no idea what they are talking about.
        • by Bigjeff5 (1143585)

          I dunno, Beowulf is my favorite mythical hero, what are we going to replace him with, Hector? Imagine a Hector Cluster of those? It just doesn't have the same ring to it.

          Hercules Cluster maybe? I dunno.

          An Achilles Cluster simply does not convey the impression you want to give in your computing environment, no matter how spectacular Achilles was.

          Calling it a "cloud" is too ephemeral, could we call it "the horde" instead?

    • More like imagine a cloud of those ;)

  • by polaris20 (893532) on Monday June 14, 2010 @11:15AM (#32565476)
    They're going to power a Ferrari out of 34.5 Vespa scooter engines.
  • by Call Me Black Cloud (616282) on Monday June 14, 2010 @11:17AM (#32565500)
    I don't know if 8 kW is a lot or a little less than a normal rack would draw. The article doesn't say...how much power would a "normal" rack consume? Isn't there a measure of computing power vs power used?
    • by ircmaxell (1117387) on Monday June 14, 2010 @11:33AM (#32565694) Homepage
      A typical 1 U server (2 processor 4 cores each) can consume between 300 and 500 watts. So 42 of them (the number in a rack) would give 336 cores at between 12kw and 21kw. As a rough number for reasonable performance servers, 300 to 500 watts per U is about what you'll find. Sure, you can find more powerful and less powerful servers, but that's a decent figure...
      • by arth1 (260657)

        A typical 1 U server (2 processor 4 cores each) can consume between 300 and 500 watts. So 42 of them (the number in a rack) would give 336 cores at between 12kw and 21kw. As a rough number for reasonable performance servers, 300 to 500 watts per U is about what you'll find. Sure, you can find more powerful and less powerful servers, but that's a decent figure...

        So, since this is an 8U at 8000W, it consumes rather a lot MORE power than 8 1U machines. At least twice as much if going with 500W.

        So, not countin

        • by jeffmeden (135043)

          A typical 1 U server (2 processor 4 cores each) can consume between 300 and 500 watts. So 42 of them (the number in a rack) would give 336 cores at between 12kw and 21kw. As a rough number for reasonable performance servers, 300 to 500 watts per U is about what you'll find. Sure, you can find more powerful and less powerful servers, but that's a decent figure...

          So, since this is an 8U at 8000W, it consumes rather a lot MORE power than 8 1U machines. At least twice as much if going with 500W.

          So, not counting the potential cooling problems, to be competitive in performance-per-watts, it would have to provide more than the power of 16 1U dual-quad-core machines, or 128 Xeon cores.
          Of that, I have my doubts.

          Eh, fail. It's 8000W for all 42U (presuming you put 4 2000W chassis in there).

          And as an aside to the first number thrown out (300 to 500w/u) that's a little high since most servers with a 500w nameplate will, at full load, only run in the 350-400w range, and that's atypical since most servers don't spend much time at 100% usage except in rare, specific applications. Building for 20KW/rack (or beyond) is on the high end right now, with most data centers still OK with 5-10KW/rack.

        • This is 10 U at 8kw. But, it has 512 processor cores at that power usage. That's 15.6 Watts / Core. Whereas a Xeon (quad core) would be 37.5 Watts / Core (for 300 Watt, for 500 Watt it would be 62 Watts / Core). Now, you also have to look at performance / core. The Xeon will be FAR better at number crunching, but the Atom should fare pretty well at IO operations (as long as they are not memory intensive). So if you're using it as a Reverse Proxy or a large load balancer (Or another highly parallelizab
    • It depends pretty substantially on what the rack is full of. You basic "small shop" rack, lazilly part-filled with 2U boxes of no particular compute density, along with a tape drive, a KVM, maybe a switch or two, is pulling well under 8kW.

      72Us of fully loaded blade chassis, fully kittted out, might draw as much as the rack above, just spinning its cooling fans...
    • by geekboybt (866398) on Monday June 14, 2010 @11:39AM (#32565786)
      That's not too hard to calculate. I usually budget 2 A for a dual-CPU 1u server. At 120 V, that's 240 W. 8 kW divided by 240 W = 33u of servers. Multiplied by 12 cores (allowing for the new X5600 series, ignoring hyperthreading on both Atom and Xeon), you get 396 total cores of standard Xeon in 8 kW. Meanwhile, they're advertising 2048 Atom cores in the same amount of power. So, the real question becomes how powerful ~5 Atom cores are compared to 1 Xeon core.
      • by timeOday (582209)

        So, the real question becomes how powerful ~5 Atom cores are compared to 1 Xeon core.

        And, how parallel is your workload.

    • by vlm (69642)

      I don't know if 8 kW is a lot or a little less than a normal rack would draw.

      how much power would a "normal" rack consume?

      Well, just as a first guess, knowing nothing about computers, you could have estimated that you can wedge several floor mounted electrical baseboard heaters into the physical space of a rack, and the servers would (probably) be running 24x7 unlike the baseboard heaters. Keeping those baseboard heaters ventilated and air conditioned is going to be moderately challenging, but doable given a reasonable budget.

      So, knowing that its marketed as a "savings" so 8 kw is going to be less than average and that an ord

    • 8kW for a normal rack is a LOT!
      The ones I know consume one bra, a bag of bird seed, and one to two men a week. ;)

  • Or it's only me who can't find it?

    "No changes to software" or something like that.... And only tons of RFC* and "funny acronyms"... What software needs no change?

    • I couldn't find the phrase "no changes to software" in TFA (no mention even of "software" in fact), but I'm guessing it would mean that since it is still an x86 based system that you can achieve power saving benefits while using any current x86 OSes, rather than say have to rewrite/recompile stuff to run on ARM processors.

  • Low power, really? (Score:3, Interesting)

    by TheRaven64 (641858) on Monday June 14, 2010 @11:20AM (#32565546) Journal

    In all of the benchmarks that I've seen, clock for clock a Core 2 gets about twice the score of an Atom, sometimes more. The Core 2 uses a bit more than twice as much power, but if you have two Atoms you also need twice as many north-bridge chips and this pushes the power usage up to over what the Core 2 will consume. The newer Xeons do even better.

    The first benchmark results I found that compared the two were PassMark benchmarks, where a 2GHz Atom scored 386 and a Intel Xeon X5680 at 3.33GHz scored 10620. The fastest Atom, the D510 at 1.66GHz, scored 662. Even if your code scales linearly, you need more than 16 of the fastest Atom that you can buy to replace one Xeon. Or, to put it another way, this 512-Atom machine is about as powerful as a 32-CPU Xeon.

    A single Atom D520 draws around 13W, so 16 of them draw 208W. The Xeon will draw 130W. Drawing under 2KW for 512 Atoms means that they probably aren't using the fastest available ones. Actually, it means that they're drawing under 4W per Atom, which means that they're probably using Z-series Atoms, getting about half the performance of the D-series ones, so you'd only need about 16 Xeons for the same performance.

    For most workloads, if the server is not busy, you'll get much better power usage from the Xeon as well. Power usage drops off dramatically when the CPU is not 100% busy. Unless you are turning individual atoms off completely, you can't scale back power usage nearly as well with the Atoms, as single processes that would not be CPU-bound on the Xeon will require an Atom core to run at full speed.

    In other words, it sounds a lot more like greenwashing than anything that's actually sensible.

    • by Jeng (926980) on Monday June 14, 2010 @11:31AM (#32565674)

      Considering how many articles were linked I don't know if you rtfa'd or not.

      The Core 2 uses a bit more than twice as much power, but if you have two Atoms you also need twice as many north-bridge chips and this pushes the power usage up to over what the Core 2 will consume.

      This is from the wired article.

      Just changing the CPU to a low power chip, though, isn't enough says SeaMicro. The trick lies in creating a new architecture that can pull all the chips together and manage their power requirements.

      "If you just replace the chips in a traditional server with Atom processors, the power consumption actually goes up," says Feldman.

      Integrating features such as storage, networking and server management into a single ASIC (application specific integrated circuit) helps manage power better, says the company. It has also virtualized the CPU input-output so those modules that would have otherwise occupied space on a board and consumed power don't anymore.

      • Which doesn't address my point at all. Great if they can reduce the number of support chips per CPU, but if you do that and reduce the number of CPUs as well then you'll save more power than if you pick the least powerful CPU possible.
        • by Jeng (926980)

          Also from the Wired article.

          The Atom-based servers target a few specific tasks performed by data centers. In the past, servers were largely used to solve a small number of complex data-based problems, says Feldman. But the internet changed this. In the internet-focused data center, the challenge is to handle millions of small tasks such as searching, mapping and viewing pages quickly, and to do this in a way that can handle unpredictable bursts of traffic.

          They are going for mass IO not computing power.

          Here is the link again. http://www.wired.com/gadgetlab/2010/06/seamicro-server-intel-atom/ [wired.com]

    • by LWATCDR (28044)

      For performance I would wonder how this would have worked with say Cortex A9 cores instead of Atom D520s.
      Also you really might have done even better with Opterons.

      So I must ask why this over Operons, Xeons, or Arm?

      • So I must ask why this over Operons, Xeons, or Arm?

        That was my first thought too. Considering some of the benchmarks that show ARM performance per watt that crushes the Atom. But according to the Wired article:

        Though SeaMicro has used Atom processors for its chipset, the company says it has designed its architecture to be flexible and support any CPU. So any low-power chip included that from ARM, which runs on most smartphones today, can become a part of SeaMicro’s system. But Atom remains the best choice for now, says Feldman. ARM processors used in cellphones consume much lower power than an Atom chip but they also cannot deliver the same kind of computing performance, he claims.

        So they say their system can handle it but they don't think ARM's have enough performance. I'd tend to doubt that and instead bet that they just haven't built their custom chip to handle the ARM yet. If they did, they could take advantage of ARM's extensive hardware power saving features. I would think it would be much cheaper too since you could have

      • Re: (Score:3, Interesting)

        by Skal Tura (595728)

        Because you are thinking serial while they are thinking parallel.

        How many simultaneous operations can do 512 atoms VS. say total 128 Xeon cores?

        What happens when single operation is extremely small, but there are extremely high volume of them?

        What happens to a CPU core while it's waiting for RAM or other I/O? Yea, that's right: It waits.

        What happens to memory IOPS when you have 512 channels versus 128 dual-channels? Yup, it's vastly higher, but not actually just twice, but quadruple (dual channel is for dua

        • by LWATCDR (28044)

          Well then again it the ops are very small they may all fit in cache on the Xeon.
          Yes having more cpus may actually have a higher performance but then how would say 2048 Cortex A9s do then? Of course you will have to use Linux on the Arm CPUS but then for a server that isn't usually an issue.

    • by LWATCDR (28044)

      According to the wired article they use a custom chip set.
      From the article
      “If you just replace the chips in a traditional server with Atom processors, the power consumption actually goes up,” says Feldman.

      Integrating features such as storage, networking and server management into a single ASIC (application specific integrated circuit) helps manage power better, says the company. It has also virtualized the CPU input-output so those modules that would have otherwise occupied space on a board and

    • by imgod2u (812837)

      For most workloads, if the server is not busy, you'll get much better power usage from the Xeon as well.

      Not really. Look at the idle power numbers on a Xeon compared to an Atom. One of the things about low power (and low-clocking) chips is that leakage is orders of magnitude less. Atom not only runs on a lower-leakage process, it also runs at a lower voltage than Xeons. So while dynamic power-per-unit-work may not be competitive in aggregate, idle power is not even close.

  • by IYagami (136831) on Monday June 14, 2010 @11:23AM (#32565562)

    SeaMicro: 512 Atom processors in 10U rack
    Dell PowerEdge R815: 4 Opteron processors, each one up to 12 cores, in 2U rack. In a 10U you can include 5 of these servers, which will bring 4 x 12 x 5 = 240 Opteron cores

    More info here [dell.com]

  • I think the main point here is trying to make a server that can do many non-intensive parallel computations. But then when I look at a GPGPU (such as the Tesla c2050), you can get the same type of performance using a 2kW server in 3U (which I have here). The Atom is ~3GFLOPS per processor making this cluster ~1500GFLOPS strong. A single C2050 has ~500GFLOPS and you can load 4 of them in a single server. nVidia's S2050 has that performance in a single U.

    • Except that you have to rewrite you application completely to be able to use the GPGPU. And it will be impossible for most applications.

    • by imgod2u (812837)

      Atom has a TDP of ~2W and I would speculate a maximum power consumption of ~4W. A C2050 has a maximum power consumption of 247W. For the same power bugdet as a C2050, you can use 60 Atoms each with its own pocket of memory and condition code handlers.

  • following the recent single-atom transistor development, an actual simple CPU that is built from 512 atoms of various elements?

  • A ad off topic I know but I need to know. What's the deal about all these hexagonal, honeycomb like venting grills that I first saw appear on IBM gear many years ago but now seems to be ubiquitous to server gear from all vendors. Is there some cool scientific reason behind it? Is it about maximizing airflow, creating nice vortexes inside the machine to spread the air, reduce noise, reduce vibrations in the chassis, weight concerns, heat spreading.. what's the deal?
    • They look cooler than little round holes or slits, and probably use a little less metal, meaning more can be recycled after stamping.
  • who thought this headline was about making a CPU out of 512 atoms?

    The scope of atomic manipulation recently could really get this crazy.

  • This seems to be the poor man's answer to Sun's Niagra/CMT kit.
  • At the low end you want a battery to drive a mobile device ideally for weeks at a time. At the high end you want to tie zillions of devices together without spending a fortune on power and A/C. Maybe the high end solution will use larger number of low-end devices, but with an overall lower TCO.

It's later than you think, the joint Russian-American space mission has already begun.

Working...