Computing took a leap forward when chipmakers started putting more than one core—or central brain—on a single chip. It was a way to make machines work harder even as they consumed less power. Just wait until a single chip can sport 80 cores.The only realistic way to use chips like this is to let each core run independent tasks, as in agent-based computing. The alternative is to attempt to break up linear tasks into parallel subtasks. That seems a losing proposition. If we have a large collection of cores running many different tasks, we will be dealing with systems that function as agent-based models. This implies lots of indeterminism and the need for lots of self-awareness, self-governance, self-protection mechanisms.
The wait won't be long. Chipmaking giant Intel (INTC) on Feb. 11 said it has successfully produced just such a chip, the size of a fingernail, capable of processing a mind-boggling 1 trillion calculations a second. The chip, which Intel claims is the fastest ever made, could start being used commercially in 'in five years, if not sooner,' Intel Chief Technology Officer Justin Rattner says. …
The main stumbling block to widespread acceptance of such chips, however, is the difficulty in writing software to take advantage of multiple cores. Even as Intel and AMD race to deliver quad-core chips in the next few months, software developers continue to struggle to write threaded applications to take advantage of just two cores. Intel's Rattner suggests the chipmaker made the announcement of the new chip early to get software developers thinking about massively multicore chips. "If we just go two, four, eight cores, we'll never get there [with software]," he says.
Monday, February 12, 2007