IBM has become well-known for its work in greening data centers and IT in general. Less well-known, though, is its work in greening the world's most powerful computers. I chatted with Dave Turek, IBM's Vice President of Deep Computing (yes, that's his real title) about IBM's work in the field, and gained some surprising insights.
First, a few facts. IBM's supercomputers are generally the greenest in the world, according to recent rankings by The Green 500 List, which ranks the 500 most energy efficient supercomputers in the world.
The most energy-efficient supercomputer in the world is an IBM supercomputer used at at the Interdisciplinary Centre for Mathematical and Computational Modelling, University of Warsaw. It's based on QS22 Blade servers, and produces more than 536 Mflops (millions of floating point operations per second) per watt of energy.
The top 20 greenest supercomputers are all built on IBM technology, and 39 of the top 50 use IBM technology as well.
Turek told me that's no accident, and that it's a result of decisions that IBM made back in 1999, when it had to make basic decisions about its next generation of supercomputers. The company was quite prescient about projecting the current energy crunch, and so decided that it would build its new supercomputers on an architecture that would be as energy-efficient as possible. So instead of basing the architecture on energy-hogging, powerful microprocessors, it instead turned to embedded processors, which are far more energy-efficient.
"Because of that," Turek says, "our supercomputers are two to three times more energy-efficient that other supercomputers with the equivalent processing power."
This is of more than academic interest --- there's a great deal of savings involved. Turek provides some eye-opening back-of-the-envelope calculations that show just how much money can be saved when using a green supercomputer.
He says that in some parts of the U.S. energy costs are 10 cents per kilowatt hour, while in some places in Europe, the costs are two and a half times that. A green supercomputer, compared to a non-green one, can save up to $2 million in energy costs a year in the U.S. and $5 million a year in Europe.
"Many people keep these systems running for a very long time," he says, "sometimes for as long as 15 years."
The result? A potential savings of $75 million in energy costs over the lifetime of a supercomputer in Europe.
The savings can be even greater in the future. Supercomputers get increasingly powerful each year, and require more energy. Ten years from now, he says, supercomputers will be 1000 times more powerful than today. If energy costs for supercomputers escalated at the same rate as their computing power, no one would be able to afford to run them. So future supercomputers will have to be even more efficient in the coming years.
What does this all mean for IT? More than you might think, he says.
"Supercomputers are the canaries in the coal mines for IT," he says. "Because the computational needs of supercomputers are so rigorous, we bump into issues before IT sees them. So when we solve energy issues for supercomputers, we're helping IT as well."