GPU supercomputer: Nvidia Tesla cards to debut in Bull system

Paris (France) – French website PC Inpact delivered juicy details about a new supercomputer that is currently designed in France. So far, we only knew that the system will use 1068 CPUs. Now we know that it will be based on 8-core Nehalem processors as well 96 GPUs, bringing the total number crunching horsepower to 300 TFLops, enough for the second position in the current Top-500 supercomputer list.

The original Bull NovaScale supercomputer setup was based on 1068 nodes, which will result in a total of 8544 processing cores for a peak floating point performance of 192 TFlops. Apparently the design was recently upgraded with 48 nodes of Nvidia Tesla S900 GPGPU cards with 96 (GT200) GPUs. The expected performance is now about 300 TFlops. In comparison, BlueGene/L, currently the world’s most powerful supercomputer, peaks at 596 TFlops an uses 212,992 processing cores. Second is JUGENE with 65,336 cores and a peak performance of 223 TFlops.

Overall, the estimate puts the GPU performance at around 103 TFlops or almost 1.1 TFlops per GPU. Whiel the performance capability of the GPUs looks impressive, especially if 96 GPUs can achieve 54% of the floating point performance of 1068 8-core Nehalem processors, the comparison isn’t completely fair. Supercomputers realy on massive memory bandwidth and we hear from scientists that GPUs just aren’t a match for these traditional systems. However, we also hear that scientists are welcoming the arrival of GPUs and their huge potential in deskside supercomputers and more traditional systems. It is generally expected that supercomputers may move into a hybrid GPU-CPU era within the next years.

The NovaScale system will be built in 2009 and we wonder whether what final performance thsi system, will be able to achieve. Petaflops computing may be just around the corner.

Whan an interesting world we live in.

Create a new thread in the US News comments forum about this subject
This thread is closed for comments
5 comments
Comment from the forums
    Your comment
  • Earthworm_Jim
    WTB editor.

    Interesting article though.
  • homerdog
    Quote:
    Apparently the design was recently upgraded with 48 nodes of Nvidia Tesla S900 GPGPU cards with 96 (GT200) GPUs ... Overall, the estimate puts the GPU performance at around 103 TFlops or almost 1.1 TFlops per GPU.

    :o
    Quote:
    Whiel the performance capability of the GPUs looks impressive, especially if 96 GPUs can achieve 54% of the floating point performance of 1068 8-core Nehalem processors, the comparison isn’t completely fair. Supercomputers realy on massive memory bandwidth and we hear from scientists that GPUs just aren’t a match for these traditional systems.

    Since when have high end GPUs been short on memory bandwidth?
  • getto301
    yes yes thats nice, but can it run crysis all maxed out is the real question! lmao jk :p