Many computer scientists say the High-Performance Linpack test used to rate the world's Top 500 supercomputers is not the best performance measurement for supercomputers. "What we're most interested in is being able to traverse the whole memory of the machine," says Sandia National Laboratory researcher Richard Murphy. He and his colleagues have developed the Graph500, a new benchmark that which rates supercomputers based on gigateps (billions of traversed edges) instead of petaflops.
By the Graph500 standard, supercomputers have actually been slowing down, according to Notre Dame University professor Peter Kogge. Over the past 15 years, each 1,000-fold increase in flops has resulted in a 10-fold decrease in accessible memory. According to the Graph500 standard, the top supercomputer would be Argonne National Laboratory's IBM Blue Gene-based Intrepid, which recorded 6.6 gigateps.
The U.S. Defense Advanced Research Projects Agency, the Department of Energy, and the National Science Foundation also have developed a new benchmark called the HPC Challenge, which tests computing power and memory accessibility.
From IEEE Spectrum
View Full Article
Abstracts Copyright © 2011 Information Inc., Bethesda, Maryland, USA