Int. J. High Perform. C. 22, 149 (2008)

Credit: PIPP, SXC

A radical redesign of supercomputers may be crucial to developing more definitive climate models, say scientists. Current climate prediction capabilities are hampered, in part, by the relatively coarse resolution of climate models and the high monetary and computational costs of running them.

Leonid Oliker at Lawrence Berkeley National Laboratory in Berkeley, California, and colleagues propose the development of a new supercomputer that borrows design principles from iPods and mobile phones rather than conventional desktop computers. The use of multiple highly efficient microprocessors dramatically reduces energy demands and increases computing speed. Building a traditional supercomputer to run the advanced climate model envisioned would cost over US$1 billion, and the computer itself would need as much power as a small city, generating a carbon footprint equivalent to the energy consumption of 100,000 people if run continuously for a year. The new climate computer, in contrast, would be built for a fraction of that cost and would use only two per cent as much energy to run the same advanced climate model.

Existing climate models are generally run using 100-kilometre or greater spatial resolutions. This supercomputer would theoretically be capable of running global simulations that resolved atmospheric, cloud and climatic processes at a kilometre scale.