Rice building Texas’ fastest academic supercomputer

BY JADE BOYD
Rice News Staff

Rice University has secured grant funding from the National Science Foundation and  Intel Corp. to build a supercomputer that will rank among the world’s fastest.

When fully operational next year, the Rice Terascale Cluster (RTC) will be approximately three times faster than any other university computer in Texas. The RTC will consist of at least 70 interconnected servers containing the powerful new Intel “Itanium” 2 processor.

Housed at Rice’s Computer and Information Technology Institute (CITI), the RTC will be the first university computer in Texas with a peak performance of 1 teraflop, or 1 trillion floating-point operations per second (FLOPs), the standard measure of supercomputer performance. The total cost for the RTC is undetermined. Funding includes $1.15 million from the NSF.

Were it operational today, the RTC would rank among the 10 fastest academic supercomputers in the country and the top 25 university computers worldwide, according to <www.top500.org>, a semi-annual ranking of the world’s top supercomputers that is compiled by researchers at the University of Tennessee and Germany’s University of Mannheim.

Currently, the fastest supercomputer in Texas, according to the list, is the University of Texas at Austin’s IBM Regatta-HPC Cluster, which has a peak performance of one-third of a teraflop.

Scientists need faster computers to tackle increasingly complex mathematical problems that would require weeks or months to compute on existing machines. For example, to precisely map the movements of every atom in a large molecule, researchers need to develop a complex mathematical model that contains thousands of variables. Such models are useful for drug designers and biomedical researchers, and a whole new scientific discipline known as bioinformatics has been created to solve this and other complex biological computations. Increasingly, research across academic disciplines requires a similar level of complex computation, and it also requires a new generation of distributed software.

“Rice faculty from disciplines as diverse as biochemistry, earth science, economics, neuroscience, computer science and political science will use RTC in their research,” said Moshe Vardi, RTC principal investigator and CITI director. “It will also be a vital tool for basic computational research aimed at better designing software that can run on hundreds or even thousands of processors simultaneously.”

Rice’s proposal for NSF funding for the RTC faced stiff competition in a process that saw awards for just one in three applicants. Rice won based on independent evaluations by reviewers who praised CITI’s expertise in high-performance computing, the interdisciplinary nature of CITI research and the caliber of the faculty involved.

Complex research already slated for the RTC includes simulations of biomolecular interactions, the physics of heavy ion collisions, simulations of Internet-based computer applications running on hundreds of computers and simulations that aim to better understand and predict international conflicts.

The RTC is slated to begin operation in early 2003. Tentative plans call for the cluster to include 70 interconnected Hewlett Packard servers, each containing four 900-megahertz, Itanium 2 processors. The cluster will have more than 500 gigabytes of RAM, and it will be linked to a 1 terabyte array of dedicated hard drives.

According to <www.top500.org>, most of the fastest supercomputers, including the world’s fastest — the 35.9-teraflop NEC Earth Simulator in Japan — and the United States’ fastest — the 7.2-teraflop ASCI White-Pacific at Lawrence Livermore National Laboratory in California — are operated by private or government-run research laboratories. See <www.top500.org/list/2002/06/> for the Top 500 list of supercomputers.