The Age of the Supercomputer

Ten years ago, Deep Blue beat a world chess champion. Now supercomputers are poised to move into the business world in a big way

by Dave Turek

On May 11, it will have been 10 years since Deep Blue became the first machine to defeat a reigning world champion chess master under tournament rules. The eight-day, six-game match with Garry Kasparov drew the world's attention. It was parodied by late-night comics and in comic strips. It became fodder for folk singers, documentary producers, and book authors. Deep Blue, a machine built by IBM (IBM), was a pop star.

Things have changed in 10 years. Today, Microsoft's (MSFT) Xbox 360 game console is more powerful than Deep Blue, and the real silicon speedsters only make headlines for the fleeting moment when they wear the title, "Fastest Supercomputer in the World."

Deep Blue was not the fastest computer in the world in 1997. On a list of the top 500 fastest computers, Deep Blue ranked a poky 388. Still, its 32 processors could handle about 11 billion calculations per second.

Ten years and four processor generations later, Blue Gene, the present fastest supercomputer and the progeny of Deep Blue, uses 131,000 processors to handle 280 trillion operations every second.

The Beauty of Virtual Models

The biggest advance isn't the technology per se. A decade since Deep Blue, the business world is now witnessing the advent of fully commercial supercomputing. The real revolution is yet to come.

Supercomputing is moving rapidly from its traditional place in academic and government research to tackle a growing list of commercial applications in business. Of the top 500 supercomputers in existence in 1997, only 161 were employed in businesses. Today, nearly half of the 500 speediest—246 systems—reside in commercial enterprises.

Two essential economic forces are propelling that movement. The first is the ever-rising expense of building and testing physical prototype designs for everything from diapers to cars. The second is the rapidly decreasing cost of perfectly modeling and testing a virtual product on a supercomputer.

In 1997 the price for the capability of processing 1 million operations per second was roughly $50. At that rate, Deep Blue's 11 billion calculations cost about $550,000.

Today, 1 million ops cost about 10 cents—or $1,100 for the same computing capacity. As a general rule, supercomputer power increases and prices fall about 40% each year. The fastest supercomputer in the world in 1997, a multimillion-dollar capital expenditure, is today's $200,000 starter kit.

Modeling Financial Risk

Experienced users of supercomputers are riding the price/performance curve to apply the machines' incredible power to areas that not so long ago would have been considered too esoteric or tangential to waste valuable resources on. Procter & Gamble (PG), for example, has become so adept and efficient in its use of supercomputers that it can afford to tweak the design of Pringles potato chips so they fly more efficiently through the production line; make diapers more eco-friendly and absorbent; and make soap suds bubblier using computer modeling and simulation techniques that until recently were the province of aerospace engineers.

Early pioneers, in industries where massive computer power has long been justified as a necessary capital expense, will continue to demand machines that push the limits of computer theory. As the cost of drilling a single well in deep water approaches $35 million, we can expect the energy industry to further expand the boundaries of computational modeling of petroleum and gas reserves. Some of the new applications that will emerge among large users lie in managing digital media and modeling financial risk.