How do you create a 1 exaFLOPS supercomputer (1000 times faster than the current leader, the 1 petaFLOPS Roadrunner supercomputer)?

By creating a grid of 4 million volunteers with 1 teraFLOPS GPU-equipped PCs, available an average of 25% of the time, says David Anderson, founder of the popular volunteer computing site known as BOINC (Berkeley Open Infrastructure for Network Computing), which he estimates could happen in 2010 — years ahead of other paradigms.*

GPU architectures have steadily become more general-purpose, and the latest models do double-precision floating-point math. In 2007, NVIDIA released a system called CUDA that allows GPUs to be programmed in the C language, , making it much easier for scientists to develop and port applications to run on GPUs.

* Based on the current speed trend, KurzweilAI.net projects that supercomputer speeds would hit 1 exaFLOPS in 2018.