PROCESSORS

Winners of the 2003 Gordon Bell Prizes, one of high performance computing's most prestigious honors, will be announced at the SC2003 conference on high performance computing and networking. The Gordon Bell Prizes are awarded each year at the annual SC conference to recognize outstanding achievement in the field of high performance computing. Gordon Bell Prizes can be awarded in three categories: special accomplishment based on innovation; peak performance based on operations per second; and a price per performance ratio measured in megaflop/s per dollar. Winners depend on the entries received; in some years a prize is not awarded in a given category. The Gordon Bell Award winners will be announced by Fred Johnson of the U.S. Department of Energy during the conference awards session starting at 1:30 p.m. Thursday, Nov. 20. Prior to the awards session, the finalists will present their work during technical sessions held on November 19: “Gordon Bell Computational Methods” at 1:30 p.m. and “Gordon Bell Performance Evaluation” at 3:30 p.m.. This year's finalists include: High Resolution Forward and Inverse Earthquake Modeling on Terascale Computers. Authors: Volkan Akcelik, Jacobo Bielak, Ioannis Epanomeritakis, Antonio Fernandez, Omar Ghattas, Eui Joong, Julio Lopez, David O'Hallaron, and Tiankai Tu (Carnegie Mellon University); George Biros (Courant Institute, New York University); and John Urbanic (Pittsburgh Supercomputing Center). For earthquake simulations to play an important role in the reduction of seismic risk, they must be capable of high resolution and high fidelity. The researchers developed earthquake simulation algorithms and tools and used them to carry out 1 Hz simulations of the 1994 Northridge earthquake in the Los Angeles Basin using 100 million grid points. IPSAP : A High-performance Parallel Finite Element Code for Large-scale Structural Analysis Based on Domain-wise Multifrontal Technique. Authors: Seung Jo Kim and Chang Sung Lee (Department of Aerospace Engineering, Seoul National Universtiy, Korea); Jeong Ho Kim, Minsu Joh, and Sangsan Lee (High Performance Computing and Networking Supercomputing Center, Korea). The researchers developed and used a domain-wise multifrontal solver, IPSAP, a code that can overcome most of the difficulties associated with large-scale parallel structural analysis, to tackle eigen problems with seven million degrees of freedom within one hour. A New Parallel Kernel-independent Fast Multipole Method. Authors: Lexing Ying, George Biros, Denis Zorin, and Harper Langston (New York University). The researchers developed an extremely efficient parallel implementation of an adaptive fast multipole algorithm that can be applied to problems such as viscous flows, fracture mechanics and screened Coulombic interactions. The researchers ran an experiment that included 1.2 billion unknowns, and achieved 1 Tflops/s peak performance and 0.7 Tflops/s sustained performance. Performance Evaluation and Tuning of GRAPE-6—Towards 40 "Teal" Tflops. Authors: Junichiro Makino and Hiroshi Daisaka (Department of Astronomy, School of Science, University of Tokyo); Eiichiro Kokubo (National Astronomical Observatory of Japan); and Toshiyuki Fukushige (Department of General System Studies, College of Arts and Sciences, University of Tokyo). The researchers benchmarked GRAPE-6, a sixth-generation special-purpose computer for gravitational many-body problems, and presented the measured performance for a few real applications with a top speed of 35.3 teraflops. A 14.6 Billion Degrees of Freedom, 5 Teraflops, 2.5 Terabyte Earthquake Simulation on the Earth Simulator. Dimitri Komatitsch, Chen Ji, and Jeroen Tromp (California Institute of Technology); and Seiji Tsuboi (Institute for Frontier Research on Earth Evolution, JAMSTEC). The researchers used 1944 processors of the Earth Simulator to model seismic wave propagation resulting from large earthquakes. The model, based on a very high-resolution mesh, incorporates wave speed and density structure, three-dimensional wave-speed and density structure, ellipticity, topography, and bathymetry. The Space Simulator: Modeling the Universe from Supernovae to Cosmology. Authors: Michael S. Warren, Chris L. Fryer, and M. Patrick Goda (LANL). Researchers developed the Space Simulator, a 294-processor Beowulf cluster with theoretical peak performance just below 1.5 teraflops that was built for less than $500,000. This is the first machine in the TOP100 to surpass Linpack price/performance of $1 per Mflop/s. SC2003 continues the 15-year Supercomputing Conference tradition of highlighting the most innovative developments in high-performance computing and networking. Bringing together scientists, engineers, researchers, educators, programmers, system administrators and managers, SC2003 in Phoenix will demonstrate how these developments are sparking new ideas and new industries, as well as rekindling older ones. The conference features the latest scientific and technical innovations from around the world, while its SC Global events will showcase achievements in the arts and sciences among dozens of remote locations. SC2003 is sponsored by the Institute of Electrical and Electronics Engineers Computer Society and by the Association for Computing Machinery's Special Interest Group on Computer Architecture. For more information, please see http://www.sc-conference.org/sc2003/.