Niwa, the National Institute of Water and Atmospheric Research, quadrupled the power of its massively parallel Cray supercomputer last week with the aid of a crane, a removed building wall panel and a bit of smart manoeuvring.
The machine now closely couples 544 processors for a total processor power of 500 Gflops (500 billion floating-point operations per second), placing it around 390th on the scale of the world’s computers. But this still underestimates its true grunt, says Uddstrom.
Closely coupled processors like in the Cray are in practice more powerful and more scalable than a cluster of equivalent processors because of the fast transmission of data from processor to processor through memory.
“It will scale linearly [in computing capacity] up to about 1000 processors,” Uddstrom says, while clusters run out of steam due to the slower communication. “They’re better suited to solving a lot of small separate problems,” he says, while the Cray will be used on a single large problem such as simulating the New Zealand climate, or the changes in fish populations.
It may also be used on some global climate modelling, but local modelling fills a gap in the science in this area, says climate scientist Jim Renwick.
“On the scale of global models, New Zealand is only three or four grid points.”