Par Lab Seminar: Moore's Law goes Multicore: The economic and strategic consequences of a fundamental change in how computers work

Lecture: Par Lab | November 27 | 1-2:30 p.m. | Soda Hall, Wozniak Lounge, 430 Soda

Speaker/Performer: Neil Thompson, Haas School of Business

Sponsor: Parallel Computing Laboratory

Par Lab Seminar: Neil Thompson - Haas School of Business, Tues Nov 27 from 1-2:30pm, Wozniak Lounge

Speaker: Neil Thompson

Title: Moore's Law goes Multicore: The economic and strategic consequences of a fundamental change in how computers work

Abstract: The changeover from clock-speed improvements to multicore computing was a profound one. It meant that, where possible, software needed to be parallelized to take advantage of chip improvements and, where not possible, software would have greatly diminished benefit from chip improvements. Evidence from benchmarks gathered by Hennessy and Patterson (2007) and for this research support this view, showing much slower improvement after the changeover to multicore.

This paper examines the impact that these changes in software performance have for firms. It does this using a survey of computer scientists on the parallelism in major software packages, and extensive data on the software usage and economic performance of a large group of Swedish firms. Initial results suggest that the impact of the changeover to multicore has had important impacts on firm productivity, with firms that rely poorly parallelized programs growing productivity significantly slower after the changeover.

Event Contact: roxana@eecs.berkeley.edu