Trip to the Edge

Friday, September 1, 2006

Turn convention on its head. That’s exactly what associate professors of computer science Doug Burger and Stephen Keckler had in mind six years ago, when they set out to design a radically new computer chip capable of performing one trillion operations per second.

Steve Keckler, Doug Burger and Kathryn McKinley. Photo by Matt Lankes

At the time, Burger and Keckler had published a paper predicting the beginning of the end for conventional microprocessor architectures. The remarkable leaps in speed that had defined the industry for decades had started tailing off. Desperate for higher performance, chipmakers tended towards greater complexity, but the designs consumed too much power and produced too much heat.

“It’s a really interesting time because the industry is at an inflection point—the old ways have stopped working and everybody knows it,” Burger says. “The industry is shifting to multicore [adding more microprocessors to single chips] as a way of buying time, but it isn’t a long range solution.”

Those trends motivated Burger and Keckler to reinvent how the underlying architectures should work and rethink how instructions could interact with each other. Their collaboration led to the development of Explicit Data Graph Execution, or EDGE architecture.

Unlike conventional architectures that process one instruction at a time, EDGE can process blocks of instructions all at once and more efficiently. Though years of development remain, Burger and Keckler believe their chip design could reach supercomputer speeds. Their goal is to reach one trillion instructions per second by 2012, a potential that convinced the Defense Advanced Research Projects Agency (DARPA) to begin funding the project in 2001.

For now, the design team is awaiting delivery of the first EDGE-based microprocessor, TRIPS (Tera-op Reliable Intelligently Adaptive Processing System). Once IBM finishes fabricating the prototype in August, they’ll be able to test it with the program developed by their colleague, professor Kathryn McKinley. She’s leading the compiler effort, translating software programs into machine code that the chip can understand.

Burger and Keckler are anxious for the chance to prove their theories in real world applications and continue work on the next generations of TRIPS processors. “Our hope is that TRIPS will revolutionize the industry and have far reaching significance beyond the academic community,” Burger says.

“Of course we’re excited, but even if everything doesn’t go exactly as planned,” Keckler adds, “the project has been a huge success. Intellectually, it’s already had a substantial influence on other researchers.”