General-purpose computation on graphics processing units (GPGPU) is the technique of using GPUs, which typically handle computation for computer graphics, to perform data-parallel computation in applications traditionally handled by CPUs.
The latest NVIDIA GPU (Tesla M2050 / M2070) can provide computing power of a CPU cluster at 1/20th the power consumption and 1/10th of the cost. A single GPU can reach the peak double precision floating point performance of 515 Gflops, and the peak single precision floating point performance of 1.03 Tflops.
