The Most Powerful PCs of 2008

About the author
Read more
This thread is closed for comments
10 comments
    Your comment
  • kelfen
    wow where do they get the software to run 8 gpu's teheheh
  • voidrunner
    multithreaded notepad
  • "multithreaded notepad"

    haha made me laugh
  • LightWeightX
    This article should be entitled "Computers of the Rich and Famous".
  • zodiacfml
    and the first place is half the price of the 2nd place machine.
  • mapesdhs
    In June/08 I talked to Joost Batenburg, one of the researchers
    who uses the FASTRA system (at the time they were looking
    into better cooling solutions and I had some suggestions).

    The important point about this system is that it does not
    involve any communication between the GPUs, ie. SLI is
    irrelevant. Indeed, the gaming 3DMark06 score for the system
    is not that high, partly because it's only really using one
    card for the test (the mbd is a Crossfire board, chosen for
    other reasons).

    Indeed, after it was first announced, some people asked why
    they didn't use 9800GTX cards instead of 9800GX2s. The GTX
    is faster for games, but not by much, whereas the tomography
    computations the researchers are doing can fully use both
    GPUs on a GX2 card, giving a virtually perfect linear scaling
    for multiple GPUs, ie. much better than a GTX card.

    Oddly enough, there *is* a commercial system that can use
    up to 8 gfx cards (ie. 16 GPUs), namely SGI's top-end VS
    system (up 8 quad-core Opterons, 8 x PCIe cards, 256GB RAM),
    which is a rebadged version of the top-end system from BOXX.
    However, FASTRA is cheaper, and they don't need much in the
    way of main-CPU power anyway so a system supporting lots of
    normal CPUs would be a waste.

    FASTRA is interesting because of how it is used, but a bit
    strange that toms would put it into the list since it's not
    a games machine.

    Btw, the researchers' previous 'device', as toms puts it,
    was a cluster of 256 dual-core Opterons.

    For full details of FASTRA, see:

    http://fastra.ua.ac.be/en/faq.html
    http://fastra.ua.ac.be/en/technicalfaq.html

    Supporting lots of gfx 'pipes' is nothing new though; SGI's
    been doing it for nearly 20 years.

    Ian.

    SGI Guru.
    mapesdhs@yahoo.com
  • Pei-chen
    Upgrade the #1 machine with four GTX 295 and it would triple the performance.
  • Wow in SA we hardly even hear about stuff like that, and why would they run a Phenom in there, a Core i7 SLAUGHTERS it in its dreams! I mean even the Phenom II is weak compared to the Core i7, even the 920 model outruns anything AMD have to offer which is extemely sad
  • mapesdhs
    Read the article and the references I mentioned. The
    processing the researchers are doing does not require a good
    main CPU (it's all done on the GPUs via CUDA). An i7 is
    totally unnecessary for their needs. Besides, they needed a
    mbd with four PCIe slots all with double-spacing - such an i7
    board does not exist.

    I wish people would read the full article and background info
    before commenting.

    And btw, four GTX295s would not triple the performance. It
    would be double at best. Check toms' specs on the various
    dual-GPU boards - no way is a single speed-reduced GTX280
    core 3X faster than a single 9800 core. If it's the number
    of SPs that matters most, then the speed increase would be
    a tad under 2X. However, heat is a problem for a build like
    this, as is driver/BIOS support to get all four cards
    recognised. The GTX295 is so new (even out yet?) I'd be
    surprised if any mbd could utilise four of them atm.

    Joost told me they wanted to get the system up and running
    quickly. Wouldn't surprise if they bought the same cards
    again for any 2nd system since they know it will work ok.

    Ian.
  • ImmortalJman
    Well, by the way the GTX 295 was out yesterday on Newegg, pretty well hidden. Here it is
    http://www.newegg.com/Product/Product.aspx?Item=N82E16814130439

    I'm very surprised how they were able to get all the GPU's working but seeming as how it's a CUDA project, I'm pretty sure that they had some good help from NVIDIA.