The Gordon Bell Prizes are awarded each year at the SC conference to recognize outstanding achievement in high performance computing. This year two teams of finalists, one supported by the Office of Basic Energy Sciences and one by the Office of Biological and Environmental Research, submitted entries based in part on computations run at NERSC.

Several members of the team that won the Gordon Bell Prize for having the fastest application in 1998 were finalists again at SC2001. Their entry, "Multi-Teraflop/s Spin Dynamics Studies of the Magnetic Structure of FeMn/Co Interfaces," achieved a maximum execution rate of 2.46 teraflop/s using 2,176 processors on NERSC's IBM SP. Their large-scale quantum mechanical simulations, involving 2,016-atom super-cell models, revealed details of the orientational configuration of the magnetic moments at the iron-manganese/cobalt interface that are unobtainable by any other means. This work is of fundamental importance in improving magnetic multi-layer computer storage and read head devices. The team included Andrew Canning from NERSC; B. Ujfalussy from the University of Tennessee, Knoxville; T. C. Schulthess, X.-G. Zhang, W. A. Shelton, D. M. C. Nicholson, and G. M. Stocks from Oak Ridge National Laboratory; Yang Wang from the Pittsburgh Supercomputer Center; and T. Dirks from IBM.

Climate models have been notably absent in previous Gordon Bell competitions due to their inability to scale to large processor counts. This year, however, Richard Loft, Stephen Thomas, and John Dennis of the National Center for Atmospheric Research were finalists with the entry "Terascale Spectral Element Dynamical Core for Atmospheric General Circulation Models (GCMs)." This team is working to speed up the integration rate of climate models by using the spectral elements method for the dynamical core; this method gives better performance on SMP architectures than the more commonly used spectral transform method. Using 2,048 processors on the NERSC IBM SP, and achieving a sustained performance of 369 Gflop/s, the NCAR team demonstrated that their GCM dynamical core can be integrated at a rate of 130 modeling years per computing day.