Parallel Computing Mainstream in APAM

Apr 15 2009

The cost of computing has plummeted by more than four orders of magnitude over the past 20 years, from a market-leading $2,500 per installed Mflop/s in 1989 to about $0.12 today, as represented by APAM’s new 2.1 Tflop/s SiCortex system. (1 Mflop/s is one million arithmetic floating point operations per second, and 1 Tflop/s is a million times that.) As a result, computer simulation has become the modality of scientific discovery and engineering design of first resort in many fields, and is used to narrow the parameter space that must ultimately be explored by generally more expensive forms of physical experimentation. This is a far cry from the mantra of computer pioneers, such as R. W. Hamming of Bell Labs, who declared that the purpose of computing was “insight, not numbers.”

APAM is therefore pleased to offer in-house cost-free access to a 1458-processor parallel computer for research and education, thanks to pooling of faculty start-up allowances and contributions from the department, Dean, and senior faculty. Beyond the new machine (in the renovated computer area in the Plasma Physics Laboratory on the first floor of Mudd) called “Amdahl” after the parallel computing pioneer from IBM, as consortium members, APAM’ers may also apply for privileged access to the supercomputer currently ranked #28 in the world: a 100 Tflop/s, 36K-processor IBM BlueGene/L located at Brookhaven National Laboratory.

Users of Amdahl or BlueGene must convert their algorithms into a form that explicitly manages the passing of data between distributed memories.  To this end, APAM post-doctoral researcher Ian Langmore, himself a relative newcomer to parallel computing, has been educating new users.

Stay up-to-date with the Columbia Engineering newsletter

* indicates required