Ditches proprietary GPGPU solution; to support DX11 and OpenCL in future
AMD has decided to stop development of its Close To Metal solution for general-purpose GPU programming on its graphics cards, according to Tom's Hardware.
Reporting on a speach by AMD Graphics Product Group CTO Raja Koduri, the company believes that the day of proprietary solutions for programming general-purpose computing on GPUs, such as Nvidia's higher-level CUDA, has passed - and from now on AMD will only support standardised solutions.
As such, it'll be focusing on DirectX 11's compute shaders and the new OpenCL standard. For those unfamiliar, OpenCL is a new language for heterogenerous data programming proposed by Apple, and currently being standardised by Khronos Group - the company behind other standards such as the open 3D asset sharing format COLLADA.
Given Intel's recently announced Larabee GPU, which rumours suggest will consist of between 16 and 24 x86 cores each with its own vector unit, it's becoming increasingly clear that unless a common format is reached there will be significant fragmentation between the three vendors. On Windows, DirectX is most likely to unify things much as it did with vertex and pixel shaders, but OpenCL hopes to do what OpenGL and OpenAL have done before it by providing a both chipset- and platform-agnostic solution, and Apple's intention to integrate the standard with the next version of OS X is likely to help boost uptake.