PROCESSORS
The Future of Really Big Computers
Supercomputers are the gleam in every scientist's eye, useful for such data-intensive research as simulating global climate or unraveling the human genome. They work by a concept known as "parallel processing," whereby multiple computer chips execute parts of a program simultaneously. The more chips, the bigger the problem the computer can handle and the faster it can do it. Over the last decade, says Thomas Sterling, a faculty associate in the Center for Advanced Computing Research at the California Institute of Technology, the performance of supercomputers has increased almost one thousand times, from less than 70 gigaflops, or one billion calculations per second, in 1993, to over 35 teraflops, or one trillion calculations per second, in 2003. And yet, many of today's largest systems often demonstrate disappointing efficiency, even though their size, cost, and power consumption continue to escalate. Current design strategies won't work for the next generation of supercomputers, making it unlikely they will be able to achieve the future potential speed of multi-petaflop computing (a quadrillion calculations per second).
On Wednesday, November 5, Sterling, a leader in the field of high-performance computer architecture, will discuss the challenges and the possible solutions for supercomputers in his talk "From PCs to Petaflops--The Future of Really Big Computers," the second of the 2003-2004 Earnest C. Watson Lecture Series at Caltech.
New research at Caltech, JPL, and other institutions, he says, is pushing the frontiers of supercomputers. By the end of the decade this work may revolutionize the way in which such computers are built and operated, and solve the problem of performance degradation. Sterling's talk will describe the very biggest computers ever built, and predict what future supercomputers will look like and what they may be able to achieve.
Sterling's lecture will take place at 8 p.m. in Beckman Auditorium, near Michigan Avenue south of Del Mar Boulevard, on Caltech's campus in Pasadena. Seating is available on a free, no-ticket-required, first-come, first-served basis. Caltech has offered the Watson Lecture Series since 1922, when it was conceived by the late Caltech physicist Earnest Watson as a way to explain science to the local community.