Forgot your password?
typodupeerror
Australia Supercomputing Science Technology

New Supercomputer Boosts Aussie SKA Telescope Bid 32

Posted by Soulskill
from the it's-not-the-size-of-your-cluster-it's-how-you-use-it dept.
angry tapir writes "Australian academic supercomputing consortium iVEC has acquired another major supercomputer, Fornax, to be based at the University of Western Australia, to further the country's ability to conduct data-intensive research. The SGI GPU-based system, also known as iVEC@UWA, is made up of 96 nodes, each containing two 6-core Intel Xeon X5650 CPUs, an NVIDIA Tesla C2050 GPU, 48 GB RAM and 7TB of storage. All up, the system has 1152 cores, 96 GPUs and an additional dedicated 500TB fabric attached storage-based global filesystem. The system is a boost to the Australian-NZ bid to host the Square Kilometer Array radio telescope."
This discussion has been archived. No new comments can be posted.

New Supercomputer Boosts Aussie SKA Telescope Bid

Comments Filter:
  • And still it can't run Crysis...
    • by Anubis350 (772791)
      Not that the c2050's drivers are optimized for gaming, but it certainly has the grunt to play games, I've done it on the one in my tower right now :-p
  • by Mick R (932337)
    should we put another GPU on the barbie for them?
    • Re: (Score:3, Informative)

      by crutchy (1949900)
      c'mon... we don't "put" anything on the barbie here. stuff is always chucked on :)
      i hope australia gets the ska. we have a factional space industry, but every bit extra helps
    • D'you think there's room, with Ken on there already?...
  • This machine is tiny compared to at least one other open supercomputer in Australia, the Vayu [nci.org.au] cluster at the National Computing Infrastructure's NAtional Facility in Canberra, which has a bit less than 12,000 cores, several petabytes of storage (tape and disk), and I believe some GPU's attached (the Xe [nci.org.au] system at the NCI NF has 16 Fermi cards).

    In the scheme of things, this probably won't help the bid much at all given its small size (and the truly astronomical amount of data the SKA will produce). But tha
  • by niktemadur (793971) on Wednesday September 28, 2011 @05:16AM (#37537656)

    Considering that the distance between 2 or more synchronized antennae becomes part of the radio telescope itself, the RadioAstron [newscientist.com] gives me chills just thinking about it, some astonishing science should come out of this bad boy.

    • by RogerWilco (99615)

      While space based VLBI is certainly cool, and some of the components in RadioAstron were developed at my company, it doesn't compare to the SKA.

      The RadioAstron has a 10 m antenna, so about 75 m2 collecting area. The SKA will have a of about 1000000 m2. (square kilometer). The baselines will be up to 6-7000 km long in some possible variants, compared to the 189000km of the Radioastron.

      So the spacial resolution might be a bit smaller (7000/189000 = 0.037), but the sensitivity is much lower (75/1000000=0.00007

  • I'm waiting for the newer model. It'z da Reggae Telescope, mon!
  • by mjpaci (33725)

    This is totally one step beyond! It's madness!

  • In other words, their new "supercomputer" still brings less to the table than all but the smallest of BOINC projects.

    • apart from decent working memory for each node, low inter-node latencies, and a filesystem

      • by RogerWilco (99615)

        The main thing actually is bandwidth.

        Radioastronomy needs a lot of CPU power, but compared to other things like the stuff that uses BOINC, it needs a lot of bandwidth.

        For example our current LOFAR system handles 200 Gb/s with a 12.000 core BlueGene and a ~3200 core, 200 machine cluster and about 6 PB of storage.

        At full resolution we would write 2 PB per 24 hours.

        With 2 million BOINC users, it would require an average stream of about 1 Mb/s per user, if each of them is online for 2.4 hours a day.

        In other wor

  • The BlueGene Stella at LOFAR central processing currently does 37 TFLops, which is about the same as this new Fornax system the article mentions (40 TFlops).

    The Stella was a big computer in 2005 in http://science.slashdot.org/story/05/05/01/2316248/When-Lofar-Meets-Stella [slashdot.org]

    Not so impressive in 2011.

    It's the difference between being #6 on the supercomputer top 500, or not even making it.

    And while LOFAR is running several other clusters besides the Stella, and manages to handle 200 GB/s, SKA is supposed to hand

If God had a beard, he'd be a UNIX programmer.

Working...