The concept of Purpose-Based Benchmarking (PBB) was proposed to DARPA and Sun by Gustafson in late 2002, and received immediate acceptance by both as the key to HPC productivity measurement. DARPA went so far as to mandate the approach in its HPCS Phase II proposal solicitation, and is working to establish a "Top 500" supercomputing list based on PBBs. This paper is probably the clearest presentation to date of Gustafson's idea, dating back to about 1990, that benchmarks must measure how well computers solve problems and not how much activity they perform.