John D. McCalpin, Ph.D

Welcome to the University of Virginia page of John McCalpin (aka "The Bandwidth Bigot", aka "Dr. Bandwidth").


Please note that I have no formal affiliation with the University of Virginia -- this account has been provided to me as a courtesy (originally sponsored by Professors Batson and Wulf, and now by Professor Skadron) to serve as a repository for the STREAM Benchmark web site and archive.

What is STREAM?

The STREAM benchmark is a simple program that has become the industry standard measure of sustained memory bandwidth for computer systems. Some history (pdf).


Who is John McCalpin?

STREAM was developed while I was an assistant professor at the University of Delaware, and remains an independent academic project.

I currently work for the Texas Advanced Computing Center at the University of Texas.

Previously I worked at AMD in Austin, TX, at IBM in Austin, TX, and SGI in Mountain View, CA. The University of Delaware was before all of those.

NEW! McCalpin's Blog on Performance NEW!

Correspondence relating to STREAM should be sent to:

A Random Selection of My Presentations

Extending my analytical models of SPEC CPU benchmark performance from two degrees of freedom (peak CPU rate plus sustained memory bandwidth) to three degrees of freedom (adding memory latency): System Balance and Application Balance in Cost/Performance Optimization presented at the "SPEC Benchmarking Joint US/Europe Colloquium" in Dresden, 2007.

A presentation on Optimum System Balance (ppt) that I originally presented at SuperComputing2004.

A somewhat vague set of hallunications entitled Toward a Grand Unified Theory of High Performance Computing (pdf) that I presented as the keynote talk at the SP-XXL conference in August 2004.

A presentation on the Memory Wall (ppt) that I gave at the University of California at San Diego/San Diego Supercomputing Center in February 2004.

A presentation on "Simple Composite Performance Metrics" (pdf) that I gave at SuperComputing2003.

I presented a slightly shorter version of this presentation (pdf) on a panel at the same conference.

A proposal (ppt) to augment the TOP500 list with STREAM benchmark data and a Simple Composite Performance Metric -- presented at SuperComputing2002. This was part of early discussions that eventually led to the development of the HPC Challenge Benchmark Suite.

A talk presented at the DOE/NSA Workshop on the Performance Characterization of Algorithms, July 17, 2001, Oakland, CA

My Keynote Address from the Third Annual Workshop on Workload Characterization, held in Austin TX, September 16, 2000.