Power Use is Major Factor in Supercomputer Growth

At the International Supercomputing Conference many of the computers are including power consumption along with their performance numbers. ComputerWorld writes the following article.

June 18, 2008 (Computerworld) Proving that size isn't everything, the Top500 List of supercomputers for the first time is looking at power efficiency.

With the IT industry increasingly looking at growing electric bills and calls for greener machines, the world's most powerful and largest systems are under scrutiny. The latest edition of the twice yearly Top500 List of supercomputers was unveiled today at the International Supercomputing Conference in Dresden, Germany. At the same time, the list authors also provided energy-efficiency calculations for many of the computers on the list.

"Power consumption is becoming one of the most important aspects of computing," said Jack Dongarra, a co-creator of the Top 500 List and a distinguished professor at the University of Tennessee. "It will be the most important driving force for supercomputing in the future. Without focusing on that, building bigger machines will be prohibitive. We're trying to understand which machines are more efficient, why they're more efficient, and understand the trends in high-performance computing."

Other interesting parts of the article.

As the industry now moves beyond the petaflop barrier toward the loftier exaflop barrier, Dongarra pointed out that increasing power efficiency becomes even more important.

"The projections for exaflop machines [show that they] will require on the order of 100 megawatts of power," said Dongarra. "It will require very special facilities to run that. The cost of running a petaflop machine at a modest electrical price will almost equal the money you pay for the computer."

He noted that engineers at the Oak Ridge National Laboratory are working with Cray Inc. to build two supercomputers -- Jaguar and Kraken. Both are in the early stages of production but are expected to surpass a petaflop. Dongarra said when they are fully functional, which is expected to happen in 2012, the power costs for both will total an estimated $32 million a year.

The list  is here.

For the first time, the TOP500 list will also provide energy efficiency calculations for many of the computing systems and will continue tracking them in consistent manner.