Slashdot Mirror

← Back to Stories (view on slashdot.org)

SAPAC Unveils New Australian Supercomputer

Posted by CowboyNeal on from the big-iron-from-down-under dept.

Sean Burford writes "The South Australian Partnership for Advanced Computing (SAPAC) has unveiled its new AU$1.7 Million supercomputer named Hydra. It is an IBM 1350 Linux cluster with 126 compute nodes (xSeries 335), 1 head node (xSeries 335), 1 storage node (xSeries 345) and 1 managment node (xSeries 345). Hydra has a peak theoretical performance of 1.2 Teraflops, and has currently benchmarked at 682 Gigaflops. The current benchmark places it in the fastest three supercomputers in Australia and equivalent to the current number 80 in the world. The cluster has a total of 258 2.4Ghz Intel Xeon processors and 258GB of RAM. SAPAC expects to achieve a benchmark closer to 700 Gigaflops with further tuning. Hydra is hosted at The University Of Adelaide, who already host a 40 node cluster of Sun e420 machines."

8 of 227 comments (clear)

  1. Is this news? by KFury · · Score: 4, Interesting

    Not to flame or troll, but considering that over 90% of the top 80 came out in the last 30 months, how big a deal is this? Third fastest computer in Australia? Sheesh.

    A computer faster than this is born every two weeks.

    --

    Kevin Fox
    1. Re:Is this news? by PerryMason · · Score: 4, Interesting

      Its really nothing huge, but you also need to consider the cost involved. AU$1.7 million is about US$1.1 million. So for about a million US$1 you could get in the top 100 supercomputers in the world.

      Looking at the latest top 500 list this would put it as the third most powerful 'self-made' system in the world. For that reason I think it deserves at least a mention and add the relatively low cost and you've got a /. story.

      --
      "I'm tired of all this 'Aren't humanity great' bullshit. We're a virus with shoes" - Bill Hicks
  2. GB vs. MB by lingqi · · Score: 3, Interesting
    ...and 258GB of RAM...

    Is it me or anyone else misread it as "256MB"?

    Actually, misreading it lead me to think about a mainframe at my college, which was an SGI with 12 processors and 512MB of memory.

    The thing is, though - when I first went in the college, we were all like "WOW that's a lot of system resources." When I got out four years later I was carrying that much memory on my laptop...

    breakneck speeds, man.

    However, regardless - (with all due respect) why is this such a big deal that australia limped to #80 on the fastest computer list? didn't other linux clusters break teraflops quite a long time ago? EarthSim was neat because it put THAT much more distance between another country and the US (and nearly nobody saw it coming) - but this seems to me hardly news, besides the possible "one of the fastest computer in australia runs linux," or something...

    --

    My life in the land of the rising sun.

  3. Why aren't they using Athlons? by leereyno · · Score: 3, Interesting

    It really is too bad they can't use Athlons.

    The per-clock performance on an Athlon is much better than what you'll get from a P4 based Xeon, and that is just on integer. When it comes to floating-point performance a lower clocked Athlon will meet or beat the performance of a higher-clocked P4.

    Right now the only SMP chipset for the Athlons is the 761, which is several years old and lacks dual-channel capability. It also requires the use of registered ECC memory. If the Athlon's had an SMP chipset comparable to the NForce2 or Intel's 775 then it would be a very different story.

    Right now the going rate on pricewatch for an Athlon 3000 is only $10 more than a 2.4 Ghz Xeon, and it would spank that Xeon on floating point which is exactly what is important for a supercomputer.

    I hope that the clustering technology they're using makes good use of SMP systems because if it doesn't then they may very well have misspent their money.

    Lee

    --
    Muslim community leaders warn of backlash from tomorrow morning's terrorist attack.
    1. Re:Why aren't they using Athlons? by sjames · · Score: 5, Interesting

      Actually, Athlon is not faster than the P4 Xeons at double precision floating point. Double precision is frequently required for scientific computing and is used in the Linpack benchmark for the top500.

      The current dual Athlon chipset is the 760MPX. The Intel i7501 is the preferred chipset for dual P4. It supports dual channel ECC DDR ram and the 533MHz FSB. These days, nobody wants non-ECC ram for a top 500 cluster. It's not that much more expensive these days compared to decently rated non-ECC ram.

      Then there's the cost of air conditioning 258 Athlons...

      I'd say they made the right decision.

    2. Re:Why aren't they using Athlons? by MetricT · · Score: 5, Interesting

      We're just a few days from bringing up a 300 processor cluster of the exact same type of computers they are using, so maybe I can shed some light. There are several reasons for picking Xeons over Athlons at the moment.

      1. If your app uses double precision floating point, and you can recompile your app using SSE2, an Intel will easily beat the AMD. AMD does scalar floating point operations faster per clock. Intel does vector flops faster. Most interesting real-world problems use vector flops.

      2. Memory bandwidth. Most chipsets can only deliver a fraction of their theoretical bandwidth. I've seen speed differences of 25% running code on identically configured machines, one having Intel E7500 and the other with a ServerWorks GC-LE (the ServerWorks smokes...) And those are *good* chipsets. I have yet to see an Athlon chipset that wasn't crap.

      3. Managability. The x335's are pretty damned slick. I *love* the built-in KVM switch and remote diagnostics. You can daisy chain north of 21 nodes together (I think 35!) and you just have one cable coming off of them.

      4. Total cost of ownership. Our previous p3 cluster was assembled (before I arrived) from Pricewatch parts. We initially experienced a 25% failure rate on memory, and spend an inordinate amount of time fixing random problems. 40 of the p3 nodes takes more than three times as much administrator time as 160 IBM x335's. Spending an extra $50,000 on good, quality parts is cheaper than hiring a competent sysadmin. Don't "efficient" yourself to death.

      Having said all that, I'm *really* looking forward to Opteron. We're getting some in a week or so. 64 bit + SSE2 support is going hard to beat.

  4. Re:Australian rules powers of 2^38B or what? by sould · · Score: 4, Interesting

    Dunno about the processors - I dont see any reason why you'd need a power of two for them. Perhaps we've got two arrays of 2^7 processors with a controller processor each (=128+1 *2 = 258)

    But I suspect as far as the ram goes that the 258 gigs is 256 - but counting 1k as 1000 instead of 1024. (or possibly 1M as 10^6 instead of 1048576)

    Haven't you noticed the difference between what a vendor says is the size of a HDD compared to how many gigs you actually get when you put it in your PC?

  5. Couple of other details not in the article by lachlancs · · Score: 3, Interesting

    The cluster is running IBM Cluster Systems Management, not Beowulf, and is using Myrinet Networking.