Chances are that when you think about supercomputing, you think about big machines (or lots of machines) all running full bore while performing complex calculations to determine weather patterns or wind-tunnel simulations. Secondarily is the assumption about power - you need lots of it, as well as ways of cooling those systems down (which requires even more power).
However, Phillip Dickens, a Computer Science professor with the University of Maine, had a very different vision in mind. Dickens applied for a grant to develop a grid computing system for the university that would provide both a supercomputing resource that would be available for everyone from researchers to school children - and wanted to do it in such a way as to not add a large energy footprint to the planet from this particular grid.
To do this, he adapted a Linux based SiCortex SC072 system to run the grid itself, chosen because the system has low energy consumption (300 W). The resulting grid (with 648 processors) is now one of the most energy efficient research grids out today.
To showcase this, Dickens choose a novel means of demonstrating just how efficient modern computers have become. At the rollout of the grid, he set up a demonstration where he had rigged up twenty bicycles into power generating harnesses, and with several buff cyclists manning the bikes, kept the grid up and running for twenty minutes as it ran through the complex calculations to simulate ice melting patterns on glaciers.
Normally, of course, the system is powered by the local electrical network, but this demonstrates that you can have supercomputing capabilities without the need for supercomputing energy costs. Perhaps it's time to hook up my laptop to my exercise bike ...