Researchers from the US Department of Energy's Lawrence Berkeley National Laboratory have suggested building a supercomputer based on low-power embedded microprocessors to improve global climate change predictions.

Using the embedded microprocessor technology used in mobile phones, iPods and other consumer electronic devices, the boffins propose a cost-effective machine for running complex computational models.

In a paper published in the May issue of the International Journal of High Performance Computing Applications, Michael Wehner and Lenny Oliker, of Berkeley Lab's Computational Research Division, and John Shalf, of the National Energy Research Scientific Computing Center, describe using the new class of supercomputers for modelling climate conditions.

A key problem that the supercomputer would address is the development of accurate cloud simulations.

Although cloud systems have been included in climate models in the past, they lack the details that could improve the accuracy of climate predictions.

Wehner, Oliker and Shalf set out to establish a practical estimate for building a supercomputer capable of creating climate models at the one-kilometre scale.

A cloud system model at this scale would provide rich details that are not available from existing models.

To develop the model, scientists would need a supercomputer that is 1,000 times more powerful than is available today, the researchers say.

They estimate that building a conventional supercomputer capable of modelling clouds at a 1km scale would cost about $1bn.

The system would also require 200 megawatts of electricity to operate, enough to power a small city of 100,000 residents.

In a paper entitled 'Towards Ultra-High Resolution models of Climate and Weather' the researchers present an alternative that would cost less to build and require less electricity to operate.

They conclude that a supercomputer using about 20 million embedded microprocessors would deliver the results and cost $75m to construct.

This "climate computer" would consume less than 4-megawatts of power and achieve a peak performance of 200 petaflops.