New Climate Supercomputer Made of “iPod Chips”


New concept may provide a powerful platform for climate science.

iPod with Earth image

Researchers at U.S. Department of Energy’s Lawrence Berkeley National Laboratory are hoping to build a supercomputer a thousand times more powerful than any currently available. That’s the sort of computational horsepower required for advanced climate modeling. Climatologists would like to simulate cloud behavior down to a single kilometer.

Building a conventional computer on this scale is possible, but such a system would literally require enough power to light a small city. There’s also the expense: an estimated $1 billion USD.

But three Berkeley Lab researchers think they have a cheaper, dramatically more efficient solution: a massively parallel array of low-power processors similar to those which run iPods, cellphones, and portable devices. Such chips are already designed to run cool on as little wattage as possible, opening up new possibilities in the development of a supercomputer which is essentially a vast network of embedded processor cores.

Whatever floats your point

In a paper published this week, the Berkeley scientists estimated an “iPod supercomputer” could be constructed for about $75 million. The proposed architecture would be optimized for small processor cores, optimized links, and the requirements of highly parallel applications.

So how many tiny little cores would such a supercomputer utilize? The target is 20 million, which would yield peak performance around 200 petaflops and only suck down about four megawatts.

New models for a new computer

Researchers want to step well beyond current climate models, applying the proposed supercomputer’s power to brand-new mathematical concepts developed at the National Center for Atmospheric Research in Boulder, Colorado. The granularity of the new models would simply be impractical on current hardware.

The Lawrence Berkeley National Laboratory has already inked an agreement with Tensilica, Inc., to adapt their Xtensa processor core as the building block of the “climate computer.”The Xtensa is thought to be about 400 times more efficient per watt in the performance of floating point operations.

Berkeley Lab is a U.S. Department of Energy national laboratory located in Berkeley, California. It conducts unclassified scientific research and is managed by the University of California.

Further Reading:

Berkeley Lab Researchers Propose a New Breed of Supercomputers for Improving Global Climate Predictions (Berkeley Lab)

There Are 5 Responses So Far. »

  1. [...] Purpose Computing Focuses on Energy Efficiency: A couple weeks back, we reported on a Berkeley research team’s plan to build a powerful climate simulator out of iPod-like [...]

  2. Is this really useful information, or just and academic exercise. If practical justification is possible, then it can be cosidered an achievemnt, but like putting people on Mars, or fusion, if there is no benefit, why spend the money?

  3. Is this really useful information, or just and academic exercise. If practical justification is possible, then it can be cosidered an achievemnt, but like putting people on Mars, or fusion, if there is no benefit, why spend the money?

  4. I suppose it depends whether you think climate research is important. I do, whether it's modeling atmospheric temperatures or forecasting hurricanes. Getting to 1 km resolution is a stunning feat — and doing it for a fraction of conventional computer costs gets my thumbs up.

  5. I suppose it depends whether you think climate research is important. I do, whether it's modeling atmospheric temperatures or forecasting hurricanes. Getting to 1 km resolution is a stunning feat — and doing it for a fraction of conventional computer costs gets my thumbs up.

Post a Response. (Want an avatar? Register at Gravatar.)