Since 1986 - Covering the Fastest Computers in the World and the People Who Run Them

Language Flags
September 15, 2006

Fastest Doesn’t Equal Best

By Peter Meade

First off, let me say I have no problem with speed. In fact, I love speed. In most cases, I equate speed with cars. I enjoy the fact my car can go 0- to-60 mph in 5.7 seconds. That's pretty fast, certainly swift enough to be keeping a constant eye out for John Law.

But at no time do I equate fastest with best, especially after being overtaken by some teenager in a hyper-turboed Civic. I feel this way too about speed as it relates to computers. Sure, I don't like to wait for computations and I relish being able to have five browsers and an equal amount of files open at the same time while still speeding along at the keyboard, but that isn't my only criterion for judging a computer's worthiness. This leads me to the bone I have to pick with some factions of the supercomputer industry. I'm sorry, but I will never believe that just because the Ferrari Enzo is the world's fastest car (0- 60 in 3.28 seconds), it is therefore the world's best. So why are many of the supercomputer faithful so equally entranced by speed as the ultimate metric for determining the best machine?

The Japanese, no strangers to fast cars, are obsessed with speed — as well as winning back the title of builder of the world's fastest supercomputer. Just as Mr. Ford learned years ago that he'd better look for Mr. Toyota coming fast in his rear view mirror, IBM must know Blue Gene/L's days as the fastest machine in the world can be nothing more than numbered.

Japan makes no secret of the fact it is aiming to develop a supercomputer that will regain the crown as the world's fastest. The country lost the cherished designation to the U.S. last year, so the Japanese government has countered by stating that it plans to develop a supercomputer that can handle in excess of one quadrillion calculations per second by as early as March 2011. That may seem like a while off, but not when speaking the hyper-speedy language of supercomputing.

If achieved, the performance would well surpass the 70.72 trillion calculations per second mark set in independent tests last year by IBM's Blue Gene/L. The world's record holder was built for the U.S. Department of Energy's National Nuclear Security Agency. Currently, Japan's fastest machine is NEC Corp.'s Earth Simulator, which boasts nearly 36 trillion calculations per second. This machine was dethroned as the world's fastest last year when it was knocked off by wares from Silicon Graphics Inc. and IBM.

The U.S. and Japan have been embroiled in spirited supercomputer competition for many years and expect there will be no end in sight to the rivalry, which much of the time takes on an exaggerated mantle of national pride.

The Japanese ministry has enlisted NEC, Hitachi Ltd., the University of Tokyo and Kyushu University to develop the much-needed technologies to make the super-swift computer a reality. Take that for true seriousness the U.S. would be hard pressed to match.

That said, I remain firm that the fastest supercomputer isn't the best. Just like cars are asked to do many things besides go fast, it's much more important to have the correct vehicle for the application at hand. One of my favorite analogies (and I have lots of them) is that while it would be fun to drive that Ferrari Enzo to the supermarket, it is not the best choice for hauling home a week's worth of groceries and such. We must not forget to factor in application, as well as other elements including reliability and price relevant to budget, when determining the best computer for the job.

But in this era, typified by the relentless pang that even instant gratification takes too long, expect the battle for supercomputer supremacy to be more of an Le Mans than a drag race.