Marching towards petascale computing

The U.S.'s race to build the first supercomputer capable of executing a thousand trillion floating point operations per second (or petaflops) continues. According to the New York Times, the National Science Foundation will award IBM a $400 million contract to build a 1 petaflop machine. The National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign will run the system on behalf of the Defense Advanced research Projects Agency.

This will be the third petaflop machine the federal government has commissioned. Last year, Oak Ridge National Laboratory announced it will get a Cray 1 petaflop computer running by 2008. IBM also has hinted that its BlueGene/L supercomputer, run by Lawrence Livermore National Laboratory, will be built out to petascale levels by 2008 as well.

A few years back, the White House and Congress were troubled by how Japan had passed the U.S. at having the most high performance computing power at its disposal, and revved up supercomputing funding, resulting the the above-mentioned machines.

But floating point operations per second is a peculiar measure of supercomputer success, given how such operations are but one requirement for many applications (and not even a major one in many cases). Presumably these machines are configured to handle a wide range of jobs beyond the task of scoring well on the Top 500 List of supercomputers.

Posted by Joab Jackson on Aug 08, 2007 at 2:33 PM