Supercomputing for the Masses

(page 2 of 2)

But creating new markets predicated on making supercomputer performance mainstream isn't a slam dunk. There's no clear path to the kind of kind of high-volume markets, measured in the hundreds of millions of units sold, that the tech industry counts on to fund its advances. And efforts to apply high-performance computing power to a broader set of problems could be stalled by a dearth of widespread programming knowhow (BusinessWeek.com, 5/2/07).

"Is this whole infatuation with performance something that has moved beyond what the vast majority of users really care about?" asks Intel's Rattner. "Are there really a set of applications that require 10, 100, 1,000 times the performance we have today? And if we have it at an attractive price point, will it drive high volumes? It's still to be determined," he says. "There are still people who question whether the volume markets are there for all this performance."

Intel is pondering these questions as it prepares to celebrate the 60th anniversary of the invention of the transistor, a mainstay of its products, at Bell Labs in 1947. Intel's most recent quad-core chips contain about 820 million transistors, which amplify electrical signals and let current flow.

Programming Advances Required to End Customization

High-performance computing has been a growth market for tech companies, even as demand for more traditional business systems has ebbed. The market for high-performance servers alone reached $10 billion in 2006, and grew 18% in the third quarter of 2007, to $3 billion, according to market researcher IDC (IDC). "The real growth in revenue has been at the bottom half of the market," as segmented by price, says Ed Turkel, a manager in HP's high-performance computing division. Distributing work across dozens of processors to speed performance "will ripple down to PC desktops as well," he says. "We're at the front of that technology curve."

Combined with advances in graphics processing and the spread of high-speed Internet connections, researchers envision more immersive online worlds with realistic graphics, and "personalized information spaces." These information spaces will feature computers that can track conversations, anticipate the needs of participants, and pull up any documents relevant to the discussion. They're also exploring systems that can quickly compile magnetic-resonance-imaging data in a doctor's office so patients don't have to wait days for results, or allow architects to see instantly how changes to a concert hall design would affect acoustics. "This is a real sea change that's happening in computing," says Thom Dunning, Reed's replacement as director of the NCSA, which lends its supercomputer center to companies and universities for research projects.

But if the computer industry hopes to apply supercomputing techniques to products designed for millions of users, it's going to have to match the performance gains in hardware with significant advances in software and programming techniques. Most software companies—and their customers—can't afford the expense of custom designs, which is the way most supercomputers are used today. "For the mass market, you can't count on customization," says the NCSA's Dunning. "It's far too expensive. We're keeping our fingers crossed that Microsoft, Intel, AMD, and other companies will provide some incentive."

Ricadela is a writer for BusinessWeek.com in Silicon Valley.