IBM supercomputers rated most energy efficient

Supercomputers from IBM are more energy efficient than supercomputing iron from rival vendors, according to latest research from The Green500.org.

Whilst the list of vendors supplying supercomputing iron is fairly limited (IBM, Cray, SGI, HP, et al) the latest Supercomputing Green 500 List has found that the top 20 most energy-efficient supercomputers in the world are built on IBM high-performance computing technology. Indeed, Big Blue holds 39 of the top 50 positions on the list.

This year's report was sponsored by server components supplier Supermicro.

It found that the most energy-efficient supercomputer in the world is an IBM machine at the Interdisciplinary Centre for Mathematical and Computational Modelling at the University of Warsaw. The system produces more than 536 Mflops (millions of floating point operations per second) per watt of energy.

IBM wasted little time in pointing out that rival iron offers much less performance per watt of energy. The first non-IBM entry on the list (from SGI) offers 240 Mflops per watt, whereas the most energy efficient system from HP offers 217 Mflops per watt.

"Modern supercomputers can no long focus only on raw performance," said David Turek, vice president of IBM's deep computing division, in a statement. "To be commercially viable these systems most also be energy efficient."

The world's fastest supercomputer, reportedly the IBM petaflop supercomputer at Los Alamos National Laboratories, is ranked ninth in energy efficiency.

There was no indication from The Green500.org as to why IBM's hardware is so much more energy efficient than rival iron. It could be down to the fact that Big Blue is using newer designs or is making use of more recent microprocessors.

IBM, HP and SGI did not respond at the time of writing.

Nevertheless, awareness of supercomputer power consumption seems to be a growing issue, especially because of the cooling problems as much as anything else. This is because as a supercomputer's nodes consume and dissipate more power, they must be spaced out and aggressively cooled.

According to Professor Wu-chun Feng, director of the Synergy Laboratory and associate professor at the Department of Computer Science and Department of Electrical & Computer Engineering at Virginia Tech, without exotic cooling facilities, overheating makes traditional supercomputers too unreliable for application scientists to use.

He points out that building exotic cooling facilities can cost as much as the supercomputer itself, and operating and maintaining the facilities costs even more.

According to Wu, there has been a 10,000-fold increase since 1992 in the performance of supercomputers running parallel scientific applications, but performance per watt has only improved 300-fold.

He points out that ten of the most powerful supercomputers on the list each require up to 10 megawatts of peak power - enough to sustain a city of 40,000. One of the fastest machines, the IBM BlueGene/L, was custom-built with low-power components, "but the system still consumes several megawatts of power. At anywhere from $200,000 (£132,859) to $1.2 million per megawatt, per year, these are hardly low-cost machines."