Usually, when we think of supercomputers, we think of large, precision-cooled rooms lined with large machines. But one the most powerful such systems is a distributed system known as Folding@Home, led by the researchers in the Department of Chemistry at Stanford University. This system was the first computer system to reach 1 petaflop (10^15 floating point operations per second) last September. The system has also been recognized by the Guiness Book of World Records as the the world’s largest distributed system.
The purpose of this system is “to understand protein folding, misfolding, and related diseases” [1]. The system invites users from around the world to contribute unused processor cycles on their personal computers or other computing systems by downloading software firmware utilities. Notably, PlayStation 3 game consoles have been included and have greatly increased the effectiveness of the effort.
How was this vision accomplished? In 2000, Vijay Pande, professor of chemistry at Stanford, decided to start the project to help in large scale computations about his research area in protein folding. Why? As the site FAQ states, “our application needs not the hundreds of processors found in modern supercomputers, but hundreds of thousands of processors” [2]. Prof. Pande has archived this vision and in 2008, 250,000 active CPUs contribute to the project.
The role of Sony’s PlayStation, with a high-end cell processor and graphics hardware designed for playing games, cannot be understated in this project. The game console’s users now make up a million of the project’s members, and more are signing up all of the time. Furthermore, due in part to the game console’s specialized hardware, its users contribute more than half on the computing power that has been collected to date [3].
The technical idea behind the project is a large number of computers connected together through a specially engineered network. Each computer solves a small part of a scientific problem–called a work unit–and then returns it to a central server. The algorithms used for this calculation were developed especially for the purpose. In particular, methods know as Implicit Solvent Models were redesigned to work on PCs with slow communications layers (the Internet) and to incorporate as many contributors as possible. The result? Problems that no one had previously solved are now performed on a regular basis. Using the power of the calculation distribution, Prof. Pande has collected data that has been used in 54 scientific papers to date. All papers, along with the raw data, is released for other researchers to use.
What can we learn from this example? In numerical computing, many “intractable” problems can be solved through the creative applications of technology. By applying such techniques, then greater scientific understanding will be expanded.

[1] http://folding.stanford.edu/
[2] http://folding.stanford.edu/English/FAQ
[3] http://www.vnunet.com/vnunet/news/2208966/folding-home-clocks-million-ps3