The United States is once again home to the world’s fastest computer. It ends a 20 month American absence from the top spot while first a Chinese and then a Japanese machine was number one.

The honor is bestowed by the TOP500 project, which has been putting together a regularly-updated supercomputer list since 1993. It measures sustained rather than peak performance.

The new number one (pictured) is dubbed Sequoia, runs IBM BlueGene/Q system, and is housed at the Department of Energy’s Lawrence Livermore National Laboratory in California. It’s designed to simulate nuclear weapon performance without the need for underground testing. The aim is to get a clearer idea of how long existing weapons can last, potentially giving enough confidence to delay having to pay for replacements.

It was a clear winner, with a performance of 16.32 petaflops, compared to the 10.51 petaflops of the former champion, the Fujitsu K of the RIKEN Advanced Institute for Computational Science in Kobe.

Crudely speaking, a petaflop is one quadrillion calculations per second. To put it in terms that are easier to picture, the BBC has a great analogy: Sequoia’s performance is equivalent to doing in one hour what it would take the combined efforts of everyone on earth using a pocket calculator 320 years to do.

While Sequoia is made up of more than 1.5 million processors, more than double that of Fujitsu, it’s actually more energy efficient than the Japanese machine, thanks partly to a water cooling system. That’s something of a comparative statement though: we’re talking 7.9 megawatts compared to 12.6 megawatts.

To show the progress of supercomputing, Sequoia can perform as many calculations in one second as the number one rated machine back in 1993 could do in three days.

Michael Papka of the Argonne National Laboratory in Illinois, home to the number three rated machine on the list, told Computerworld that the rapid increases were no longer largely driven by Moore’s Law. Instead he explained that most of the difference now is made by the way the processors are organized, particularly with the use of parallel processing.