Can multi-core chips deliver superlinear speedups?

Moving your code to a multi-core processor can yield a real performance boost, though the speedup is typically less than linear. For instance, moving a parallelizable algorithm from a single-core processor to a quad-core processor might give you a 3.95x performance increase, but not 4x. And it certainly couldn’t give you more than 4x.

Or could it? In a recent edition of Dr. Dobb’s, Microsoft’s Herb Sutter argues that developers can, in some circumstances, use “P cores to do work more than P times faster.” Read the article here.

No comments: