Researchers at U.S. Department of Energy’s Lawrence Berkeley National Laboratory and National Energy Research Scientific Computing Center have proposed building a new class of supercomputers based on 20 million semi-custom ASICs (embedded microprocessors) designed specifically for modeling climate conditions.

To model clouds with a 1-km (.62 miles) level of detail would require at least 10 petaflops (quadrillions of floating-point operations per second) of sustained performance, according to a statement by Berkeley Lab.

The proposed “climate computer” would enable the petascale computing required for a detailed model without the high cost and extreme power requirements of supercomputers based on conventional general-purpose microprocessors. It would deliver a sustained speed of 10 petaflops (200 petaflops peak) and cost $75 million to construct. It would use less than 4 megawatts of power.

In constrast, it has been estimated that a 10-petaflop, quad-core Opteron-based system would cost $600 million and use about 50 megawatts. The corresponding Blue Gene/P system would cost $2.6 billion and use 20 megawatts.

The climate computer would have 100 times higher computing speed/dollars/watt than the Opteron-based system and 173 times higher than the Blue Gene/P system.

See also:

HPCWire

Researchers Propose a New Breed of Supercomputers for Improving Global Climate Predictions

Towards Ultra-High Resolution Models of Climate and Weather

Climate scientists call for their own ‘Manhattan Project’