|
AMD and NVIDIA work their own API for GPGPU : stream and CUDA respectively. Situation is aggravated since both producers use their own physical engine : havok and PhysX, which will be based on API for general purpose calculations on GPU. Therefore, programmers have some issues, since the support of two API will cost additional time and means. OpenCL in this case in the only common language : an open language for calculations, released by khronos group : AMD, NVIDIA, Intel, Sun microsystems, Apple, Sony computer entertainment and others.
Recently, NVIDIA officially reported about new driver with OpenCL support . Thus , unfortunately, only participants in OpenCL early program access , can use this driver after filling the form on NVIDIA site . OpenCL API is based on own NVIDIA API for general purpose calculations on GPU CUDA.
From one side, this approach is completely justified, since CUDA support already more than 100 million video cards throughout the world. From other sides, the addition program between video cards and OpenCL hardly will positively affect the speed. However, the difference in performance can prove to be completely small, and NVIDIA will ensure support to the API new within the more compressed periods, which during the first introduction stage is important. Related Products :
|