Green 500 list ranks supercomputers

Nov 23, 2007 7:23 AM
Filed under Hardware

Researchers at Virginia Tech have released the first Green 500 list which ranks the world's supercomputers by performance-per-watt rather than pure processing power.

The list of the 500 most energy efficient supercomputers around the world is a derivative of the Top 500 list, a bi-annual report of the top performing supercomputer sites.

The project was spearheaded by Kirk Cameron and Wu Feng, associate professors at Virginia Tech's computer science department.

"The list is meant to encourage people to develop systems that use power efficiently," said Cameron.

"If you do that, you decrease costs, which are really high, between $800,000 and $1m a year per megawatt. It's ridiculously expensive. You're looking at $1m to $4m a year to run a mega system."

Virginia Tech said that the Green 500 will help usher in a new era where supercomputers can be compared by performance-per-watt, rather than just computational power.

This inaugural list uses the measured power consumed during a Linpack run, if submitted, or peak power consumption of the supercomputer otherwise.

"While the selection of any power-performance metric will be controversial, we opt for 'flops per watt' given that it has already become a widely used metric in the community," says the Green 500 website.

This method is under review by the researchers, and Feng hopes that the process will "evolve over time to ensure accuracy and more closely reflect energy efficiency in the fast-paced, ever-changing, high-performance community" .

The top of the list is completely dominated by IBM's Blue Gene supercomputers, with 26 of the top 27 deployments.

Caroline Roberts, deep computer sales manager in the UK at IBM, said that energy efficiency was a key design feature of the Blue Gene architecture.

"Since IBM sold the first Blue Gene supercomputer in 2004, it has come along in leaps and bounds," she said.

Roberts explained that using a large number of relatively low powered CPUs (850MHz compared to 2GHz used in most other supercomputers) meant that the Blue Gene machines are "sipping electricity but still producing massive computational power".

Roberts concluded that "floppage-per-watt" will become an increasingly important metric for supercomputer deployments as the pressure of green issues and costs of powering these systems continues to increase.