Slashdot Mirror


Japanese Supercomputer K Hits 10.51 Petaflops

coondoggie writes "The Japanese supercomputer ranked #1 on the Top 500 fastest supercomputers broke its own record this week by hitting 10 quadrillion calculations per second (10.51 petaflops), according to its operators, Fujitsu and Riken.
The supercomputer 'K' consists of 864 racks, comprising a total of 88,128 interconnected CPUs and has a theoretical calculation speed of 11.28 petaflops, the companies said."

125 comments

  1. And the answer was... by qualityassurancedept · · Score: 2, Funny

    42

    --
    if your life is such a big joke then why should I care?
    1. Re:And the answer was... by Tastecicles · · Score: 1

      infinity minus 1

      --
      Operation Guillotine is in effect.
    2. Re:And the answer was... by Anonymous Coward · · Score: 0

      We already knew that. It was supposed to be working on the question.

    3. Re:And the answer was... by deains · · Score: 1

      They forgot to carry the 4. Luckily they didn't reverse the polarity as well, or it may have meant the end of the universe as we know it.

    4. Re:And the answer was... by Anonymous Coward · · Score: 0

      "69, dudes!"

    5. Re:And the answer was... by Anonymous Coward · · Score: 0

      Actually,

    6. Re:And the answer was... by ZigiSamblak · · Score: 1

      I thought the whole point was to calculate the question to that answer?

    7. Re:And the answer was... by Anonymous Coward · · Score: 0

      IBM'S A2 16 64-bit cores 64-threads manufactured at 45nm running 1.4GHz all under 60 watts. Theirs future growth at 32n and 22nm. two generations left before hitting the brick-wall, which Intel is at now.

  2. Significant advance . . . by StefanJ · · Score: 1

    Fully boots Windows in under three minutes!

    1. Re:Significant advance . . . by Anonymous Coward · · Score: 0

      Microsoft windows vista "whopping great supercomputer" edition?

    2. Re:Significant advance . . . by qualityassurancedept · · Score: 1

      Really? because we keep getting Authentication Error 0xc004f050

      --
      if your life is such a big joke then why should I care?
    3. Re:Significant advance . . . by ooshna · · Score: 1

      I can cold boot to desktop in under a min with Win7 maybe you need to upgrade your windows 98 or at least uninstall that horrible Bonzi Buddy. I fucking hate that purple monkey he still haunts my dreams.

    4. Re:Significant advance . . . by Anonymous Coward · · Score: 0

      But its using server bios, so it still takes half an hour to get to loading Windows.

    5. Re:Significant advance . . . by Anonymous Coward · · Score: 0

      Highly doubtful. Even if it booted windows at all, it probably has hundreds of GB of ram (per node) and would still likely need hours to POST.

    6. Re:Significant advance . . . by jd · · Score: 1

      You've got to remember that booting the OS doesn't mean that the login or MIC mechanisms actually work, it just means they're running.

      --
      It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
    7. Re:Significant advance . . . by jd · · Score: 1

      The only way I can think you can boot to desktop on Win7 cold is if you mean liquid nitrogen cold with severe overclocking.

      --
      It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
    8. Re:Significant advance . . . by Anonymous Coward · · Score: 0

      On 88,128 Sparc CPUs?

    9. Re:Significant advance . . . by Pseudonym+Authority · · Score: 1

      Cold boot: Start the computer after power has been disconnected for a period of time.
      Warm boot: A simple reset.

    10. Re:Significant advance . . . by jd · · Score: 1

      Duh. I know what a cold boot means. But there isn't a PC that can cold boot inside of 1 minute using either a regular BIOS or EFI to a Windows desktop in under a minute unless it's overclocked.

      --
      It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
    11. Re:Significant advance . . . by Bengie · · Score: 1

      BUILD showed off a laptop that cold booted Win8 to desktop in 5-8 seconds. I thought that was pretty good.

    12. Re:Significant advance . . . by f()rK()_Bomb · · Score: 1

      i can boot to windows 7 from cold in 30seconds. 15 seconds of that is the bios setting up. SSD's are fast.

      --
      "The space elevator will be built about 50 years after everyone stops laughing." - Arthur C. Clarke ~1980
    13. Re:Significant advance . . . by Anonymous Coward · · Score: 0

      2001 called. They want their Win98 PC back. Damn, you're so old-hat. Get with the times man!

    14. Re:Significant advance . . . by Anonymous Coward · · Score: 0

      Fully boots Windows in under three minutes!

      Now THAT was funny!

    15. Re:Significant advance . . . by ooshna · · Score: 1

      Turning on fast boot in the bios and having a SSD is all you need.

    16. Re:Significant advance . . . by DigiShaman · · Score: 2

      Even if booting off the fastest SSD available, booting into any version of Windows will take at least 20 seconds. Dropping in a CPU with infinity processing power will not reduce this. That's because during the POST process, there are wait states timed against the RTC conducting all sorts of hardware polling to enure everything connected is alive. It's not a bug, it's a feature. A standardized process. You can optimize BIOS settings however by turning off unused hardware, features, and SATA ports to decrease POST times however. But again, at minimum, POST times will never be instant for an IBM based PC. Also, Windows performs similar hardware polling against the RTC at bootup as well.

      It should be noted that an Apple MacBook with an SSD boots almost instantly. But that should come to no surprise because Apple owns both the hardware and OS. They can optimize as they wish.

      --
      Life is not for the lazy.
    17. Re:Significant advance . . . by Tastecicles · · Score: 1

      Toshiba L755D-U stock everything, Win7HP: 57sec from cold. After that: virtual machines.

      OSX: 3m40s
      SuSE 11.4: 1m14s
      XPSP3: 23s
      Win2K: 1m24s

      Timed with this reply box open.

      My netbook (stock EeePC 1008HA) boots XP Home in just over a minute.

      --
      Operation Guillotine is in effect.
    18. Re:Significant advance . . . by jd · · Score: 1

      It takes longer than 15 secs for Windows to start up everything. (Booting to a usable state != booting to the point where something is on the screen. The claim was to boot to a desktop - which means all all initial services up and no further initialization being performed.)

      --
      It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
    19. Re:Significant advance . . . by Cyberax · · Score: 1

      Old story.

      http://lwn.net/Articles/299483/ - Linux booted in 5 seconds to usable desktop. Back in 2008.
      http://linux.slashdot.org/story/11/01/13/2248207/embedded-linux-1-second-cold-boot-to-qt - boot to QT in 1 second.

    20. Re:Significant advance . . . by blackicye · · Score: 1

      It should be noted that an Apple MacBook with an SSD boots almost instantly. But that should come to no surprise because Apple owns both the hardware and OS. They can optimize as they wish.

      The 13" i7 / 4GB / 256GB SSD Macbook Air takes as long to boot OSX Lion as it does to boot Windows 7.

    21. Re:Significant advance . . . by DigiShaman · · Score: 1

      Depends on the PC. VMs of Win7 and Server 2008 R2 are near instant. Depending on the PC though, Win7 will boot almost as fast as Lion for sure. But, on some PCs, the POST process takes far longer than the OS bootup process itself.

      --
      Life is not for the lazy.
    22. Re:Significant advance . . . by jd · · Score: 1

      Yeah, but a VM image isn't a cold boot. A plain VM image is essentially a warm boot (a lot of the system is already initialized) and most VMs are quite capable of handling VM images that are partially into a boot (since the first phase of the bootstrap really doesn't do anything that's important to a VM).

      If you're going to consider VMs and other such hacks, then Coreboot + Linux + image of a ramdisk in Flash would give you a 3 second bootup time to a console, just not to a GUI desktop. With a few additional hacks, you can probably make an image of an already-started X server that can be loaded in to avoid having that start up. And so on. But the more hacks and shortcuts you take, the less you can justify calling that a cold boot to desktop. Lukewarm, maybe, since parts are "cold boot", but the rest gains performance by being warm boot in all but name and not by actually being fast from cold.

      --
      It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
    23. Re:Significant advance . . . by DigiShaman · · Score: 1

      Just so we are on the same page here...

      Cold boot = When hardware is first powered on engaging a lengthy (but thorough) POST followed by the bootstrapping of the OS.
      Warm boot = Followed by a restart of the system through the OS or a hardware reset button. Hardware POST time is reduced due to certain hardware checks being omitted prior to the bootstrapping of the OS.
      Hot boot = Starting or rebooting an OS after the POST process. For example in Windows 98, you could perform a hot boot by holding down the shift key and selecting "restart".

      Technically, starting a VM from an original shutdown state is a cold boot even if it is emulated and certain POST checks have been omitted. But yes, from a functional standpoint, I agree, it's more of a warm boot if anything. As for booting from hibernation within any OS, I don't count the OS as booting up as it's just in a state of resuming. Nothing special there.

      --
      Life is not for the lazy.
    24. Re:Significant advance . . . by A12m0v · · Score: 1

      I wasn't aware Windows still ran on SPARC.

      --
      GENERATION 25: The first time you see this, copy it into your sig on any forum and add 1 to the generation.
    25. Re:Significant advance . . . by Cimexus · · Score: 1

      Huh? My PC takes less than 20 seconds from cold boot to usable desktop, and every component in it bar one is ~4 years old. 11 seconds for the BIOS and AHCI crap, and 9 seconds actual boot (broken down into 8 seconds Win logo, 1 second loading to desktop). No exaggeration - just timed it to make sure.

      The 'one component' that is not 4 years old is a Corsair Force SSD, which of course is the main reason it boots that quickly. But even with the old rotating hard drive it was well under a minute. If your PC takes a minute to boot up, something is wrong...

    26. Re:Significant advance . . . by Cimexus · · Score: 1

      Not always. My PC (with SSD) takes 9 seconds from the end of the BIOS, to on the desktop, usable, with hard drive activity completely finished and stopped. I have turned off one or two extraneous services (crap like Adobe Updater/Apple Updater/Google Updater/Office Preload etc), but other than that haven't done anything special to optimise. SSDs are very fast, especially for that 'post boot churning' stuff which is a lot of random IO.

    27. Re:Significant advance . . . by f()rK()_Bomb · · Score: 1

      it is fully usable at that time, hard disk has stopped accessing, the desktop appears after 10secs, 15 to fully usable. I can click chrome and it opens in about a second including about 30 tabs, your really underestimating the speed increase of an ssd. It is truly mind blowing fast, nothing I've ever upgraded in a pc has made as big a performance improvement.

      --
      "The space elevator will be built about 50 years after everyone stops laughing." - Arthur C. Clarke ~1980
    28. Re:Significant advance . . . by larpon · · Score: 1

      No - it's "superKomputer" a KDE project that finally let you reach the full potential of the KDE software suite.

    29. Re:Significant advance . . . by CSMoran · · Score: 1

      But there isn't a PC that can cold boot inside of 1 minute using either a regular BIOS or EFI to a Windows desktop in under a minute unless it's overclocked.

      You've just made my PC disappear.

      --
      Every end has half a stick.
    30. Re:Significant advance . . . by badkarmadayaccount · · Score: 1

      coreboot and linux allow asychronus hardware init, and paralelization. Any ideas on bootstrapping a random OS on pre-initilized bare metal?

      --
      I know tobacco is bad for you, so I smoke weed with crack.
  3. Great by bytesex · · Score: 1

    You can play a wicked game of Space Invaders on it !

    --
    Religion is what happens when nature strikes and groupthink goes wrong.
    1. Re:Great by badboy_tw2002 · · Score: 1

      How about a nice game of chess instead?

    2. Re:Great by Anonymous Coward · · Score: 0

      Space Invaders on 11+ petaflops? Nah.

      I'd set it up to mine Bitcoin...

    3. Re:Great by jd · · Score: 1

      No thanks. Blew up four worlds and sent a fifth tumbling into a black hole after GNU chess accidentally crossed pipes with Galactic Thermonuclear War IV, the Sequel.

      --
      It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
    4. Re:Great by AmberBlackCat · · Score: 1

      But you have to hit the turbo button first...

    5. Re:Great by Anonymous Coward · · Score: 0

      Raytraced Quake(1) would be the only obvious purpose.

      / aliquis too tired to login.

  4. Hentai by Anonymous Coward · · Score: 0

    Finally enough power to render 3D tentacle porn in real-time!

  5. Skynet is bigger by CurryCamel · · Score: 1

    consists of 864 racks, comprising a total of 88,128 interconnected CPUs

    Where goes the border between a supercomputer and a cluster?

    1. Re:Skynet is bigger by Plasmaphysiker · · Score: 3, Informative

      consists of 864 racks, comprising a total of 88,128 interconnected CPUs

      Where goes the border between a supercomputer and a cluster?

      Communication time. Trying to run a massively parallelized plasma physics simulation on a mere cluster is essentially a waste of time. The scaling is terrible.

    2. Re:Skynet is bigger by CurryCamel · · Score: 1

      Decrease communication time in a cluster, and you get a supercomputer?

    3. Re:Skynet is bigger by jd · · Score: 3, Informative

      Technically, a cluster can be a supercomputer if it is tightly-coupled, which basically means high bandwidth, low latency and as little overlap on the fabric as possible. (ie: 88,128 PCs linked via the Internet could be considered a grid but it would not be considered a supercomputer. The same number of PCs in a server room using a hundred or so switches, with each switch stuffed to the gills, would be considered a regular cluster. The same PCs in the same room using high-end switches linked as a Fat Tree, Butterfly or - ideally - a hypercube topology would be considered a supercomputer. The same PCs in the Cloud would be considered a torrential downpour.)

      The problem is ultimately, as Plasmaphysiker says, communication time. From a technical standpoint you can just as easily say "a supercomputer is any computer that can mimic or better a vector processor's overall performance for the same compute power". Ok, maybe not as easily as it's longer to say, but it comes to the same thing.

      --
      It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
    4. Re:Skynet is bigger by jd · · Score: 1

      Basically, yes. The distinction is whether a system is tightly-coupled or loosely-coupled.

      --
      It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
    5. Re:Skynet is bigger by Anonymous Coward · · Score: 0

      These days, yes. A supercomputer can be pretty much a tightly bound cluster with an absurdly fast interconnect (Infiniband and friends). It doesn't stop being a cluster, btw, it can be both a cluster and supercomputer, though some small supercomputers might still be shared memory machines.

      Back in the day (80s/early 90s), there used to be a slightly stricter definition of supercomputer, to do with "superconnection" (the property of having way, way, more networking than you need, basically, think of every point being connected to every other point with a full-bandwidth point to point link), but it would be absurd not to call the fastest computing system in the world a supercomputer just because it's a linux cluster.

       

    6. Re:Skynet is bigger by jd · · Score: 1

      A linux cluster using a hypercube topology of infiniband connections would give you a fully-connected system with full-bandwidth between any two points. A butterfly network will do that but in a slightly smaller subset of cases. A fat tree would give a much smaller subset, but it would still be superconnected. A fat tree using 10 gig ethernet wouldn't have the raw bandwidth for a superconnection but it could still be considered a supercomputer (just).

      --
      It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
    7. Re:Skynet is bigger by Anonymous Coward · · Score: 0

      All supercomputers in the top500 are superclusters.
      If i recall correctly Fujitsu's EnterpriseSPARC M9000 still holds the record for the most FLOPS from a single-board system, but that's considered mainframe class. There really aren't any supercomputers anymore, in the sense of a single machine.

      And communication time isn't really the difference between a cluster and a bigger (super) cluster. You'll find quite a lot of system on the top 500, using generic Gig-E as their interconnect. You'll find some using Gig-E that overpower others using infiniband. It's been reduced to a numbers game. Some push out more FLOPs by throwing more cores at it (Jaguar, RoadRunner, etc) others do it by throwing fewer, much more powerful, and uch more efficient cores at it (Riken, JAXA, EarthSimulator 2), but both types have MASSIVE ammounts of sockets and cores, that's pretty much it. You can cluster the racks in a single cabinet together and you have a cluster, cluster 800 racks together, you get a supercluster, it's all about scale.

    8. Re:Skynet is bigger by TheRaven64 · · Score: 1

      Aside from performance, supercomputers generally have a single system image, rather than an OS install on each node and some middleware to handle job distribution. They also usually implement distributed shared memory in hardware, so they appear to have a flat address space and will move data around between nodes depending on which ones are accessing it. They may also have coprocessors for things like MPI messaging. And, as the other posters said, they have very fast and very low latency interconnects.

      --
      I am TheRaven on Soylent News
  6. Getting boring - add more CPUs ( & now GPUs)! by Anonymous Coward · · Score: 0

    Whats so big in this ? Sounds like every now and then someone adds another 10,000 new chips and they have a new world record holding super computer.

  7. The real test... by Anonymous Coward · · Score: 0

    Spock: "Computer. This is a class one priority directive. Compute, to the last digit, the value of Pi."

    1. Re:The real test... by AliasMarlowe · · Score: 2

      Spock: "Computer. This is a class one priority directive. Compute, to the last digit, the value of Pi."

      Computer: "The answer is 10, base Pi."

      --
      Those who can make you believe absurdities can make you commit atrocities. - Voltaire
    2. Re:The real test... by jonahbron · · Score: 1

      Shouldn't that be "1, base Pi"?

    3. Re:The real test... by CurryCamel · · Score: 2

      1 base Pi == 1 base 10 == 1 base 2 == 1 base N

    4. Re:The real test... by Anonymous Coward · · Score: 1

      Real artificial intelligence - "No. I'm not falling for that one."

    5. Re:The real test... by Anonymous Coward · · Score: 0

      Probably not. 2 in base 2: 10. 3 in base 3: 10. 10 in base 10: 10. You get the idea.

  8. More proof that deficits don't matter by Anonymous Coward · · Score: 0

    Japan's very high (currently something like 200%) debt-to-gdp ratio has been sustainable because they invest in innovation. USA, take note!!

  9. What about what really matters? by Anonymous Coward · · Score: 0

    What does it get on SunSpider?

  10. imagine ... by heitikender · · Score: 2

    ... a Beowulf cluster of those!

  11. Obligatory... by Anonymous Coward · · Score: 0

    Crysis, can it play it?

    1. Re:Obligatory... by Lanteran · · Score: 1

      Screw crysis. Let's load dwarf fortress up on this bitch.

      --
      "People don't want to learn linux" hasn't been a valid excuse since '03.
    2. Re:Obligatory... by Ambvai · · Score: 1

      Maybe this system can finally process The Great Adamantine Space Elevator...

    3. Re:Obligatory... by Lanteran · · Score: 1

      Which abuses cave in mechanisms to pump magma to space where it falls down and boils away the oceans. At *pinky to lip* one million frames per second.

      --
      "People don't want to learn linux" hasn't been a valid excuse since '03.
  12. Pretty Interesting, But.. by Nukedoom · · Score: 1

    I wonder how that would compare to the combined computational power of every smartphone, laptop, and desktop computer around the world.

    1. Re:Pretty Interesting, But.. by Anonymous Coward · · Score: 0

      7 Billion ARM processors is equal to 50 core i7 Intel processors or 3 PowerPC A2.

    2. Re:Pretty Interesting, But.. by Surt · · Score: 1

      http://www.xyster.net/blog/?p=40
      Claims an iphone 3g can do 20Mflops double precision linpack.
      Assuming there are about 100M handsets with that level of performance:
      2000 Million (10^6), Million(10^6) flops = 2 * 10^15 = 2 petaflops, and that's just the smartphones. The laptops and desktops would both perform better, and have a much higher count.

      --
      "Who is the Journal of Quantum Physics going to believe?" --Stephen Hawking
  13. Does anyone have... by badzilla · · Score: 1

    A car analogy? Or how may libraries of congress / football fields?

    Seriously I doubt 10.51 petaflops means anything to anyone except a small coterie of supercomputer nerds.

    --
    "Don't belong. Never join. Think for yourself. Peace." V.Stone, Microsoft Corporation
    1. Re:Does anyone have... by Anonymous Coward · · Score: 0

      the common cpu you can find in a desktop goes from 20 to 60 GFLOPS.
      so there are two orders of magnitude between your desktop and this computer.

    2. Re:Does anyone have... by Ant+P. · · Score: 1

      A mid-range GPU pulls about 10 gigaflops. This is a million times that.

    3. Re:Does anyone have... by Ruie · · Score: 2

      A car analogy? Or how may libraries of congress / football fields?

      Seriously I doubt 10.51 petaflops means anything to anyone except a small coterie of supercomputer nerds.

      That's why I read Slashdot.

    4. Re:Does anyone have... by Surt · · Score: 1

      It's enough to accurately simulate the meaningful chemistry of about 1/100% of a human brain, in real time.

      --
      "Who is the Journal of Quantum Physics going to believe?" --Stephen Hawking
    5. Re:Does anyone have... by Surt · · Score: 1

      Ugh. Please mod parent down.
      To provide a more accurate and specific comparison:
      An i7 2600K can get around 130 GFLOPS.
      10 Petaflops == 10,000 Teraflops == 10,000,000 GFLOPS
      So that is around five orders of magnitude between your desktop and this computer. Not two.

      --
      "Who is the Journal of Quantum Physics going to believe?" --Stephen Hawking
    6. Re:Does anyone have... by qualityassurancedept · · Score: 1

      as for the car analogy... you can drive a Prius around the Earth 75 times on the amount of electricity it took to produce that result.

      --
      if your life is such a big joke then why should I care?
    7. Re:Does anyone have... by Nethemas+the+Great · · Score: 1

      I haven't had time to ponder any analogies, I was too busy pondering a Beowolf cluster of these things...

      --
      Two of my imaginary friends reproduced once ... with negative results.
  14. And, I imagine.. by Anonymous Coward · · Score: 0

    It would probably need it's own power plant, as well.

  15. Why don't they turn it up to eleven? by VMaN · · Score: 1

    Nigel Tufnel: The numbers all go to eleven. Look, right across the board, eleven, eleven, eleven and...
    Marty DiBergi: Oh, I see. And most amps go up to ten?
    Nigel Tufnel: Exactly.
    Marty DiBergi: Does that mean it's louder? Is it any louder?
    Nigel Tufnel: Well, it's one louder, isn't it? It's not ten. You see, most blokes, you know, will be playing at ten. You're on ten here, all the way up, all the way up, all the way up, you're on ten on your guitar. Where can you go from there? Where?
    Marty DiBergi: I don't know.
    Nigel Tufnel: Nowhere. Exactly. What we do is, if we need that extra push over the cliff, you know what we do?
    Marty DiBergi: Put it up to eleven.
    Nigel Tufnel: Eleven. Exactly. One louder.
    Marty DiBergi: Why don't you just make ten louder and make ten be the top number and make that a little louder?
    Nigel Tufnel: [pause] These go to eleven.

  16. Re:Getting boring - add more CPUs ( & now GPUs by jd · · Score: 1

    Computers don't scale linearly. Amdahl's Law. The reason home computers don't go beyond 16 cores is that even getting a 16-way SMP is a horrifically difficult problem. If you built a 32-core machine it would run SLOWER than a 16-core one because of all the overheads (locking on the bus, scheduling, interprocessor communications, stuff like that). You can't just add cores and expect a faster machine. You have to put in an enormous amount of time and effort to engineer the design and you really have to do so almost from scratch each time.

    --
    It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)
  17. Riken by Anonymous Coward · · Score: 0

    Gotta love Japanese names, they have/come up with so many cool ones.

  18. K machine technology by Required+Snark · · Score: 3, Informative
    The K supercomputer is build on SPARC technology.

    The system is still under construction and is scheduled to enter full service in November 2012 with 864 cabinets. As of the November 2011 TOP500 list, it uses 68,544 2.0GHz 8-core SPARC64 VIIIfx processors packed in 672 cabinets, for a total of 548,352 cores, manufactured by Fujitsu with 45 nm CMOS process technology. Each cabinet contains 96 compute nodes in addition to 6 IO nodes. Each compute node contains a single processor and 16 GB of memory. Its water cooling system minimizes failure rate and power consumption.

    The K uses a proprietary six-dimensional torus network interconnect called Tofu, and a Tofu-optimized Message Passing Interface based on the open-source Open MPI library. Users can create application programs adapted to either a one-, two-, or three-dimensional torus network.

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

    IBM has the Sequoia system coming on line in 2012 and it is also targeted at the 20 Petaflop range. It will be significantly more power efficient at 3000 Mflops/watt, three times lower then the K system

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

    --
    Why is Snark Required?
    1. Re:K machine technology by RicktheBrick · · Score: 1

      With so much power one would think that it could have already discovered fusion, cure to aids, heart disease and cancer. At 10,000 trillion flops per second, that would mean about 864 million trillion flops per day or about 26 billion trillion flops per month or about 312 billion trillion flops in a year.so invent something already!

    2. Re:K machine technology by Jeremy+Erwin · · Score: 1

      Be patient. Transcomputational problems take time.

    3. Re:K machine technology by Ruie · · Score: 1

      With so much power one would think that it could have already discovered fusion [...]

      No, the number of cores needs to increase by another two orders of magnitude before we achieve ignition.

    4. Re:K machine technology by Anonymous Coward · · Score: 0

      Fujitsu, eh? FM Towns SUPERMARTY.

    5. Re:K machine technology by Anonymous Coward · · Score: 0

      IBM's PowerPC A2 16 64-bit cores each core capable of four threads, manufactured at 45nm running 1.4GHz using less then 60 watts. WOW! imagine if we could get a thinkpad with say 4 cores and 16 threads (60 watts / 16 = 3.75 watts) at 3.75 watts now that is performance.

    6. Re:K machine technology by Surt · · Score: 1

      They've invented nuclear weapons maintenance, isn't that enough for you?

      --
      "Who is the Journal of Quantum Physics going to believe?" --Stephen Hawking
    7. Re:K machine technology by A12m0v · · Score: 1

      No one would buy it because nothing would run on it.

      --
      GENERATION 25: The first time you see this, copy it into your sig on any forum and add 1 to the generation.
  19. kid's stuff by Anonymous Coward · · Score: 0

    We need some inputs from all the conspicuous by their silence quantum computing nerds.

  20. Gee... by Anonymous Coward · · Score: 0

    Do you think it could handle a game of freecell? That takes an awful lot of cpu power...

  21. Re:Getting boring - add more CPUs ( & now GPUs by stewartjm · · Score: 1

    Relatively speaking 48 cores on a single board has been "affordable" since Magny Cours launched a bit around a year ago. And the new G34 Bulldozers, can give you 64 cores using the same motherboards. Even if you discount the huge 4 socket boards, the 2 socket G34 boards will give you 24-32 cores.

    So 16 cores is certainly not some sort of upper bound at the moment. The only thing keeping it out of sub 1k desktop computers is price. And that'll come down in a die shrink or 2. No fundamental new design required.

    Almost all of the hardware going into the Top500 clusters these days is standard off the shelf components. So, I'd tend to agree with the OP, it is kind of ho hum to add 10k new cores to an existing cluster, or to build yet another cluster with a nearly identical architecture to 100 others that came before it. It just requires money, and the will to spend it.

  22. Re:Getting boring - add more CPUs ( & now GPUs by Ruie · · Score: 1

    Whats so big in this ? Sounds like every now and then someone adds another 10,000 new chips and they have a new world record holding super computer.

    Interconnect. Adding more nodes has a linear cost increase, but connecting them all with low-latency links is difficult. It is easy enough to buy a 48-port switch, but a 1000 port switch with low latency is a completely different story, and if you need 10000 ports you are in fully custom territory - it can easily cost a lot more than all the nodes combined.

  23. All that computing power will be put to good use by Anonymous Coward · · Score: 0

    ...rendering HQ tentacle rape porn.

  24. Real-world uses! by GrandCow · · Score: 3, Interesting

    How many bitcoins per hour is that?

    --
    "Well kids, you tried your best, and you failed. The lesson is, never try." -Homer Simpson
    1. Re:Real-world uses! by Anonymous Coward · · Score: 0

      At the current difficulty, it's 30 Bitcoins per hour. Not much, ain't it...

    2. Re:Real-world uses! by Anonymous Coward · · Score: 1

      According to bitcoinwatch.com, the computing power is around 10% of the bitcoin network.
      145 (+/- a few) blocks are solved every 24 hours. Each block is 50 bitcoins, plus any transfer fees, but I'll ignore those.
      50*145/24 = 302 bitcoins (and change) per hour. 10 % of that would be ~30 bitcoins per hour.
      At current rates, that's around 88 dollars per hour.
      Most, if not all, of that money would probably go to pay the electric bill.

    3. Re:Real-world uses! by jovius · · Score: 1

      Thanks to you a fuse just blew in Bernanke's head.

  25. Stupid Computers by Anonymous Coward · · Score: 0

    ..don't invent things. Rather, it is smart people using them as Fast Slide Rules. We computer people tend to overestimate what we can achieve with our contraptions. The Russians proved that symbolic solutions by human brains could achieve similar things as the numerical solutions here in the west. Too often, computers are just complex heat generators.

  26. Difficult to believe... by The+Master+Control+P · · Score: 2

    Actual specs : 68500 Sparc64s, each with 8 cores. So every core can put away between 5 and 10 double-precision calculations every single cycle?

    1. Re:Difficult to believe... by Anonymous Coward · · Score: 0

      Intel processors are 4 issue and with AVX extensions can perform up to 8 double-precision operations per instruction for a total of 32 double precision calculations per cycle.

    2. Re:Difficult to believe... by Anonymous Coward · · Score: 0

      It's called SIMD, Luke. http://en.wikipedia.org/wiki/SIMD

    3. Re:Difficult to believe... by Anonymous Coward · · Score: 0

      it's the usual linpack bullshit - this is why people are trying to move to useful benchmarks like HPC challenge, graph500, etc.

      first trick is that linpack (simple 'schoolboy' matrix math) involves big long chains of 'a*b + c*d + e*f +....'. Many CPUs offer a single instruction 'fma' that computes both one * and one +. This is a single instruction, and the overhead of the extra '+' is small, but people count every such instruction as two flops, instead of 1.1 or so.

      second trick is that linpack is a truly tiny benchmark, a handwritten assembler implementation will likely fit on your screen even in a 10pt font. this makes it
      easy to workaround any peculiarities of a machine, peculiarities that would bite any real application in the ass...

      third trick is that linpack has about as regular memory acess patterns as you could ever wish for, so even a computer with a fairly poor memory hierarchy will do just great so long as it has enough bandwidth, which brings me to....

      the last trick is that the linpack benchmark is 'adjustable' - except for the so-called 'Nhalf' test (which most fake supercomputers (like "K") conveniently forget to run), the linpack benchmark used by top500 is not actually a repeatable well-defined test in the normal sense of the word, but more of a 'guideline'. in particular, you are free to adjust the size of the benchmark as you see fit. in practice, what this means is that people run tests that are ridiculously large (for example, "K" spent more than a day to solve a single matrix equation Ax=b) so that any inefficiencies in the system are amortized over very large amounts of simple "plus and times" arithmetic.

      tl;dr: yes, every core can issue that many double-precision ops per cycle, but only in completely artificial conditions.

    4. Re:Difficult to believe... by Attila+the+Bun · · Score: 1

      The SPARC64 VIIIfx can carry out 8 floating-point instructions per core per cycle.

      88,128 cpus x 8 cores x 2.0e9 cycles/s x 8 flops/cycle = 11.28 petaflops maximum theoretical speed.

    5. Re:Difficult to believe... by Anonymous Coward · · Score: 0

      Maybe one core has multiple floating point units.

  27. Re:Getting boring - add more CPUs ( & now GPUs by Anonymous Coward · · Score: 0

    IBM's PowerPC A2; 16 64-bit cores and 64 threads all using less then 80 watts manufactured at 45nm running at 1.4Ghz. IBM's PowerPC has future performance growth at 32nm, 28nm and 22nm. Intel's has already reached there limit 22nm will bring more slow cores; 8 cores each running at 800Mhz, performance increase none, new factor gotta have it!

  28. Re:Getting boring - add more CPUs ( & now GPUs by Anonymous Coward · · Score: 0

    This one's more powerful than the next three on the list combined. the leap is HUGE. they went from 1 pflop, to 2, to 10. What's impressive is that its output is nearly 5 times that of Jaguar, on just a little less than half the cores and about 25% less clock. It's just a shame they haven't gotten it fully optimized yet, JAXA was a lot more efficient on the previous incarnation of the same processor, but maybe that can be attributed to Solaris on Sparc (JAXA) vs Linux on Sparc (RIKEN).

  29. How is it used? by chebucto · · Score: 1

    I've always wondered how supercomputer time is rationed. How much does computer time on these things cost? How is the cost calculated? Is time divided up something like how it's done on a large telescope, where the controlling organization get proposals from scientists, then divvies up the computer's available time according to what's been accepted? Do they multi-task (run more than one scientists' program at one time)? Does the computer run at top power (10pf) at all times, or does the resource usage go up and down? And lastly, how hard is it to write programs to run on these things? Do the scientists do it themselves, and if so, do the people who run the supercomputer audit the code before it runs?

    Do the programs that run on supercomputers ever crash the system?

    --
    The English word fart is one of the oldest words in the English vocabulary.
    1. Re:How is it used? by bryan1945 · · Score: 1

      For the very little I know, when you rent time, you request a certain amount of capacity. So more than one thing can run at a time. I would imagine it would be hard to write something for a massively paralleled system. That's all I got.

      --
      Vote monkeys into Congress. They are cheaper and more trustworthy.
    2. Re:How is it used? by joib · · Score: 1

      How much does computer time on these things cost? How is the cost calculated? Is time divided up something like how it's done on a large telescope, where the controlling organization get proposals from scientists, then divvies up the computer's available time according to what's been accepted?

      On the supercomputer centers I'm familiar with, scientists write proposals which are evaluated by some kind of scientific steering committee which meets regularly (say, once per month), and gives out a certain amount of cpu-hours depending on the application.

      Do they multi-task (run more than one scientists' program at one time)?

      Yes. Typically the users write batch scripts requesting the amount of resources their job needs. E.g. "512 cores with at least 2 GB RAM/core, max runtime 3 days", and then they submit the batch job to a queue. At some point when there are enough free resources in the system, the batch scheduler launches the job. When the job finishes (or during its runtime) the usage is then subtracted from the quota they were awarded in the application process.

      Does the computer run at top power (10pf) at all times, or does the resource usage go up and down?

      Usually all functioning nodes are running and available for use, yes. Typically load is around 80-90% of maximum, due to scheduling inefficiencies etc. (e.g. a large parallel job needs to wait until there are enough idle cores before it can start, and so forth).

      And lastly, how hard is it to write programs to run on these things? Do the scientists do it themselves, and if so, do the people who run the supercomputer audit the code before it runs?

      Pretty tricky. Usually they use the MPI library. The programs are either written by the scientists themselves, or by other scientists working in the same field. The supercomputing center typically doesn't audit code, but may require the user to submit scalability benchmarks before allowing the user to submit large jobs. For some popular applications the supercomputing center may maintain a version themselves (so each user doesn't need to recompile it) and provide some more or less rudimentary support.

    3. Re:How is it used? by Anonymous Coward · · Score: 0

      You can't crash the entire system because they typically carve off a "domain" for you, which is effectively a big VM. It's not hard to write the kind of parallel code that runs on supercomputers -- Fortran and C++ both scale up pretty well using MPI -- but it's wickedly hard to squeeze the utmost out of those architectures when you're aiming for the absolute max of performance.

      Crashing the entire cluster is unlikely; they typically give you a "domain", which is effectively like a big VM. Billing depends very much on the organization renting out time. You might lease a dedicated domain that you pay for even when it's idle, or pay by the cpu-hour.

  30. Yeah, it can be unpleasant by PopeRatzo · · Score: 0

    I get petaflops sometimes when I eat at Super Taco Burrito down on Jackson at Halsted. Man, the mega-super is really tasty but you suffer later.

    With Super Taco Burrito and all the gyros places on the corner, that's ground zero for intestinal distress. But something keeps pulling me back there.

    Now what were we talking about?

    Oh yeah, is it Spring Back and Fall Forward or the other way around? Damn, now I got a taste for one a those mega-super burritos and an order of guac. I think they're open til midnight. Oh boy, it's gonna be one a those days tomorrow...

    --
    You are welcome on my lawn.
  31. Jaguar Will Become The Fastest Soon! by Diggester · · Score: 1

    Cray Inc. is working on updating its Jaguar supercomputer and that will enable 20-petaflop peak performance. ;) Therefore, we could soon see Jaguar as the fastest supercomputer in the world

  32. What for? by Anonymous Coward · · Score: 0

    Who has any code that can take advantage of this? Someone ought to see if Ray Kurzweil thinks he can program the singularity into this beast. Or else set up a plausible holodeck.

  33. Next by glorybe · · Score: 0

    Now the Japanese will miniaturise the monster, make it the size of a cell phone that uses very little power, market to people in the US and Microsoft will sue just because Microsoft always sues people.

  34. What discoveries are attributed to supercomputers by Anonymous Coward · · Score: 0

    What have we discovered recently that we couldn't, if we didn't have a supercomputer?

  35. #1 by mqduck · · Score: 1

    The Japanese supercomputer ranked #1 on the Top 500 fastest supercomputers

    Rumor has it that it's also #1 in the top 2,342 fastest supercomputers.

    --
    Property is theft.
    1. Re:#1 by RivenAleem · · Score: 1

      It's also #1 in the top 1 fastest supercomputers. What was your point, because I don't get it.

    2. Re:#1 by Nethemas+the+Great · · Score: 1

      Simple, he performed some useful math for us to gain perspective. You did not... :P

      --
      Two of my imaginary friends reproduced once ... with negative results.
  36. Re:What discoveries are attributed to supercompute by Chrisq · · Score: 1

    What have we discovered recently that we couldn't, if we didn't have a supercomputer?

    Weather forecasts are now 0.002% more accurate