In 1995, a good computer chip had a clock speed of about 100 megahertz. Seven years later, in 2002, a good computer chip had a clock speed of about three gigahertz — a 30-fold increase. And now, seven years later, a good computer chip has a clock speed of ... still about three gigahertz.
Four or five years ago, chip makers realized that they couldn't make chips go much faster, so they adopted a new strategy for increasing computers' power: putting multiple "cores," or processing units, on each chip.
In theory, a chip with two cores working in parallel can accomplish twice as much as a chip with one core. But software developers tend to see their programs as lists of sequential instructions, and they've had trouble breaking up those instructions in ways that take advantage of parallel processing.
Professor of Computer Science and Engineering Saman Amarasinghe and his colleagues at MIT's Computer Science and Artificial Intelligence Lab are giving them a hand.
In the past, computer science researchers had hoped that sequential programs could be converted into parallel programs automatically. "I spent a good part of my life trying to do that," says Amarasinghe. But Amarasinghe has now come to the conclusion that "if you want to get parallel performance, you have to start writing parallel code." At a high level, he believes, programmers will need to specify which tasks performed by their programs can run concurrently.
From MIT News Office
View Full Article
No entries found