Petaflops and Megawatts

July 13, 2012

Power efficiency in supercomputersCongratulations to the engineers at IBM, who have taken the USA to the top of the Top 500 list of supercomputers around the world. The new computer, named Sequoia, is located at the US Department of Energy’s Lawrence Livermore National Laboratory, and uses a mind-boggling 1.5 million processor cores to achieve 16.32 petaflops (trillion floating point operations per second).

The first list was published in 1993, and was headed by the CM-5/1024, which was designed by a company called Thinking Machines. If supercomputers had followed Moore’s law, doubling in performance every 18 months, then we’d expect Sequoia to be about 6500 times faster. In fact Sequoia is a mind-blowing 273930 times faster, delivering performance that we wouldn’t expect for another 8 years if supercomputers just followed Moore’s law.

Power efficiency doesn’t scale in the same way as Moore’s law: it’s one of the areas of electronics design that can move quite slowly. But in the world of megawatt-guzzling supercomputers it’s important. Really important! As a power engineer, I was delighted to see the focus on energy efficiency in the announcement of Sequoia hitting the top spot.
The IBM system consumes 7.9 megawatts – producing just over 2 petaflops per megawatt. Whilst this is a huge amount, it’s only about 60% of the power used by Fujitsu’s K computer, the previous holder of the number one position that delivered less than one petaflop per megawatt.

Although liquid cooling has been a major factor in the improved efficiency, the technology used in the converters that power Sequoia is a major advance on previous systems. Whilst the 1.6x increase in processing performance was impressive, the 2.4x improvement in power efficiency is a huge jump. Congratulations to the power engineers on the Sequoia team for such an awesome achievement!


Comments are closed.

Find out more about our Cool-Power Buck Regulators subscribe to vicor newsletter Contact Us

Get Connected