Supercomputers, the world’s fastest computers, aren’t just for scientists anymore: They have made big inroads into solving problems that touch our everyday lives.
Procter & Gamble uses them to design sturdier plastic laundry-soap bottles. The Riverside Casino in Iowa has one to manage thousands of electronic slot machines and keep track of millions of bets. The Italian police use them to analyze clues at murder scenes.
But as these massive systems become more affordable for both companies and researchers with their increased computing capability, scientists are still seeking new ways to speed them up, so they can better predict hurricanes or design more precise drugs.
At the world’s largest annual gathering of supercomputing geeks in Tampa, Fla., last week, companies discussed different ways in which they hope to reach the next holy grail – supercomputers that run at the speed of a petaflop, 1,000 times faster than the current champs.
“It’s a milestone,” said Dave Turek, vice president of Deep Computing at IBM.
A petaflop is equivalent to one quadrillion operations per second, or 1,000 teraflops. This is roughly as fast as 1 million personal computers, or 1 million times the speed of the fastest desktop PC.
Currently, the industry’s fastest machines are measured in teraflops, or 1 trillion floating-point operations per second.
Last week, the Blue Gene/L system designed by IBM for Lawrence Livermore was named the world’s fastest supercomputer in the Top 500 list, reaching a sustained performance of 280.6 teraflops.
One of the challenges is that most supercomputers today are using cheaper commodity chips, such as Intel’s Xeon or Advanced Micro Devices’ Opteron. These off-the-shelf chips that make up the brains of supercomputers are hitting walls in terms of how much additional performance they can sustain without generating too much heat that can slow the machines down.
So scientists and computer designers are looking at myriad ways and a host of exotic technologies to improve system performance. And many are looking at using more costly, proprietary technologies, with less reliance on the commodity chips from Intel and Advanced Micro Devices that are clustered together in racks to provide performance gains.
One area of experimentation is through the unusual use of dedicated computer chips, amid a cluster of hundreds of commodity chips.
The dedicated chips are used to accelerate a specific application or part of a program that will speed up the overall operation, especially if that one application or mathematical function is bogging down a system’s total speed.
A range of different types of chips are being used for this function. Some are referred to as accelerator chips.
“The payoff can be large,” said Addison Snell, an analyst with IDC.
