GPGPU is an acronym for the field of computer science known as "General Purpose computing on the Graphics Processing Unit (GPU)"
GPGPU is an acronym for the field of computer science known as "General Purpose computing on the Graphics Processing Unit (GPU)". The two biggest manufacturers of GPUs are NVIDIA and AMD, although Intel has recently been moving in this direction with the Haswell APUs . There are two popular frameworks for GPGPU - NVidia's CUDA, which is only supported on its own hardware, and OpenCL developed by the Khronos Group. The latter is a consortium including all of AMD, NVidia, Intel, Apple and others, but the OpenCL standard is only half-heartedly supported by NVidia - creating a partial reflection of the rivalry among GPU manufacturers in the rivalry of programming frameworks.
The attractiveness of using GPUs for other tasks largely stems from the parallel processing capabilities of many modern graphics cards. Some cards can have thousands of streams processing similar data at incredible rates.
In the past, CPUs first emulated threading/multiple data streams through interpolation of processing tasks. Over time, we gained multiple cores with multiple threads. Now video cards house a number of GPUs, hosting many more threads or streams than many CPUs, and extremely fast memory integrated together. This huge increase of threads in execution is achieved thanks to the technique SIMD which stands for Single Instruction Multiple Data. This makes an environment uniquely suited for heavy computational loads that are able to undergo parallelization. Furthermore this technique also marks one of main differences between GPUs and CPUs as they are doing best what they were designed for.
More information at http://en.wikipedia.org/wiki/GPGPU
