GPGPU

GPGPU

(General Purpose computation on GPUs) Using a graphics processing unit (GPU) for general-purpose parallel processing applications rather than rendering images for the screen. For fast results, applications such as sorting, matrix algebra, image processing and physical modeling require multiple sets of data to be processed in parallel. GPUs are also used in desktop computers for improved voice, face and gesture recognition.

The GPU functions as a coprocessor with its own memory that processes many threads simultaneously. For more information, visit www.gpgpu.org. See GPU, CUDA, OpenCL, DirectCompute, PhysX and AMD Fusion.

A Desktop GPGPU Powerhouse
This GV100 GPU from NVIDIA is a PCI Express card that performs 7.4 trillion 64-bit floating point operations per second (7.4 TFLOPS). Its Tensor performance reaches 118.5 TFLOPS (see TensorFlow). (Image courtesy of NVIDIA Corporation.)