Stanford University will soon power up a new GPU-powered supercomputer for computational biology. Developed in partnership with Scalable Informatics and Xtreme Compute Technologies, the system will comprise 100 Tesla C2050 GPUs in a heterogeneous cluster to be dubbed “numbercruncher”. The supercomputer will support research initiatives including computational biology, molecular dynamics, structural biology and chemical biology simulation.
According to Scalable Informatics CEO Joe Landman, the most interesting thing about this system is that is uses 4x NVidia Fermi C2050’s (100 GPUs in all), in an AMD compute node interconnected with InfiniBand.
Normally one would use the more expensive M20x0’s in OEM-designed nodes, but the XCT chassis allows us to use the C20x0 series. Not only that, we can use it for far more than GPUs … anything on a PCIe bus, say a bunch of Flash and GPUs. We are estimating an at least 15TF machine (on Linpack) taking the various efficiencies into account.