Wednesday, December 10, 2008

OpenCL faster computers with the help of GPU

The specifications of the programming environment (API) OpenCL (Open Computing Language) were published today. It is an open standard that allows use free to use any kind of chips for all sorts of calculations.

Hitherto reserved to the processing of graphics, massively parallel architecture of a graphics card is much more suited to some calculations than a central processor (CPU). The encoding of a video or a simulated meteorological phenomenon are calculated for example a much more effective in conjunction with dozens of processors flow of a graphics card that series with a few hearts CPU.

GPGPU technology, which is to delegate to the graphics card some calculations usually borne by the central processor, have emerged from among AMD and Nvidia have respectively launched CUDA and ATI stream. OpenCL is in turn able to benefit from any kind of chips, as chip signal processing (DSP), and not just graphics cards. It has the advantage of being a standard multi-platform open to all developers can access without payment of royalties, while competing technologies CUDA and ATI Stream require a specific development applications.

A collaboration of some thirty players in the computer, including AMD, Intel and Nvidia, developing OpenCL was initiated by Apple before being assigned to Khronos Group, the consortium maintains for years already OpenGL, cousin of dedicated graphics OpenCL. It will be an integral part of Snow Leopard, the next major version of Mac OS X, which has accelerated its development, which took only 6 months. Nvidia followed by the announcement and has already announced that its graphics cards CUDA take as provided natively support OpenCL. AMD has meanwhile made no announcement, but ATI Stream technology is based on OpenCL, ATI Radeon will probably be compatible.

Microsoft finally, who had already ousted OpenGL to its own technology Direct3D, always alone with DirectX should nevertheless expanded in its next version of support compute shaders, which fulfill the same objective as the GPGPU.

No comments: