Keynote talk challenges in chip multiprocessor memory systems

Abstract

The semiconductor industry appears on the brink of an arms race, competing to see which company can cram the most cores on a single die. Yet early entries are hardly well-balanced, general-purpose computers: Sun's Niagara has feeble floating-point performance and IBM's Cell processor is a thinly disguised GPU. Worse, no company has announced anything that will help address the real problem: programming the multithreaded applications needed to exploit the ample computational resources. This talk will discuss several challenges facing the memory system designers of emerging computation-rich chip multiprocessors.