Supercomputers: Ten million billion and counting

IN 1961 John F. Kennedy declared that America would put a man on the moon and the space race was officially on. No one ever formally declared a supercomputing race, but Mike Papka, who heads the Argonne Leadership Computing Facility at Argonne National Laboratory near Chicago, points to the debut of Japan’s Earth Simulator, which went online in 2002, as the field’s Sputnik moment. After decades of perceived American dominance, he recounts, Japan come out of the blue with the number one machine.

Every November and June, an independent organisation re-evaluates the 500 most powerful known machines in the world and ranks them at In recent years China and Japan nabbed most of the top five spots in a field where America once hogged the top ten. Japan’s K Computer, built by Fujitsu, currently rules the roost with 10.51 petaflops (computing power is measured in so-called floating-point operations per second, or flops; a petaflops is a million billion flops). China’s Tianhe-1A comes in second and a Cray computer at Oak Ridge laboratory in Tennessee sits pretty at number three. But a reshuffle is imminent: in the next five years, China plans to build 17 supercomputing centres with machines whizzing at a petaflops or more.

This year, however, America is limbering up for a comeback. Three of its national labs are being spruced up. Argonne National Laboratory’s IBM Blue Gene/Q supercomputer, christened Mira, will go online at Dr Papka’s outfit sometime in the second half of 2012. Meanwhile, the Lawrence Livermore National Laboratory in California will get a 20-petaflops BlueGene machine, Sequoia. Oak Ridge’s new hybrid CPU/GPU machine should clock between 15 and 20 petaflops.

Mira, capable of 10 petaflops, would rank in the top five if it ran today, and will be 20 times faster than Intrepid, Argonne’s current Blue Gene/P supercomputer (which will keep humming alongside Mira). It consists of 48 racks weighing 2 tonnes each. Its total oomph is toughly equivalent to that of 60m iPads. It is also energy efficient, at least as supercomputers go, crunching two billion calculations for every watt consumed.

Together with Sequoia, Mira is the latest in IBM’s Blue Gene range, the result of collaboration between the computer giant, Argonne and Livermore labs. It builds on the existing software designed for the older Blue Genes, which Argonne and 16 other research teams across the country have been tweaking on prototype hardware. As a consequence, Mira will be ready to crunch numbers from the day it is fired up.

Much of the crunching will be scientific in nature. Mira’s processing muscle will enable researchers to model exploding stars, turbulent airflow over aircraft wings and wind-turbine blades, or new materials to make better batteries, among others. But Dr Papka’s pride and joy is just a stepping stone to even more mind-boggling performance. Argonne is also home to the Exascale Technology and Computing Institute whose boss, Peter Beckmann, would like to see America invest more into developing the next generation of supercomputers: ones capable of reaching exaflops, or a billion billion calculations per second.

In February the Department of Energy, which oversees the Argonne, Livermore and Oak Ridge labs, will set out how it plans to broach the so-called exascale by the end of the decade. One goal is an exaflop computer which would use less than 20 megwatts of power, making it several orders of magnitude more efficient than Mira. The hope is that Congressmen’s minds will be blown by the vast strings of zeros in promised performance stats. And that they won’t be discouraged by the price tag, itself likely to feature a fair number of noughts.

Article source:


Powered by Facebook Comments

Leave a Reply

Your email address will not be published. Required fields are marked *