CAPS successfully scales storm prediction on 1,328 Cray T3E-900

Eagan, 12 November 97 Researchers from the Center for Analysis and Prediction of Storms (CAPS) at the University of Oklahoma recently used a 1,328 processor CRAY T3E-900 supercomputer to achieve record results on tests they hope will improve both the speed and accuracy of severe weather forecasting. The system is the largest ever made by Silicon Graphics/Cray and the world's fastest production-class computer.

The researchers of CAPS used the CRAY T3E-900 system to aid their development of the Advanced Regional Prediction System (ARPS). This is a computer model that uses supercomputers and high-resolution data from Doppler radars to predict severe storms up to nine hours in advance for specific county and city locations.

Using a very high spatial resolution, ARPS increases the accuracy and warning time of severe storms. This task requires a powerful computer: a single nine-hour forecast involves the solution of one trillion complicated equations.

The ultimate goal is to accurately predict severe weather systems up to nine hours in advance. As part of its record sustained performance levels, researchers from CAPS reached 36 gigaflops, including I/O, using 1,024 compute processors. The remaining processors were used on internal system requirements. The CRAY T3E-900 showed that the code can run about three times faster than original tests indicated, for a total speed of about 100 gigaflops on 1,024 nodes.

Sandra Wermer