Has any reason to programming with nvidia Cuda Toolkit on nvidia graphics card instead ATI SDK ? If yes, why? (better behavior, adaptation ? anything better or the same?)
And is it better to programming on python(pyopencl(OK), SDK (?) ) or C/C++ for display and work with image processing ? (Maybe another topic)
There is some confusion here. Choosing one OpenCL implementation or another is not like choosing a C++ compiler. Each OpenCL implementation is tied (connected) to a specific type of hardware and you can’t use an it unless your system has that hardware.
Only if you have an x86 processor (from AMD or Intel) you can either use the AMD or the Intel OpenCL implementations. Only the AMD one is mature so far.
Only if you have an NVidia GPU you can use the NVidia OpenCL implementation.
Only if you have an AMD/ATI GPU you can use the AMD/ATI OpenCL implementation.
Only if you have an IBM BladeCenter (unlikely) then you can use IBM’s OpenCL implementation.
This means than in some cases you may be able to use more than one implementation. For example, if you have a PC with an Intel x86 processor and an NVidia GPU then you can use both Intel’s and NVidia’s OpenCL drivers, but the Intel one can only run OpenCL in the CPU and the NVidia one can only run code in the GPU.
I hope this makes things clear.
P.S. IHVs, sorry if I have misrepresented the capabilities of your implementation.
I hope this makes things clear -> Of course, better than the beginning, really thanks.
So if I understood, with my nvidia GeForce GTX 2xx, I should use NVidia’s CUDA toolkit for OpenCL implementation, right? I can’t use ATI SDK?
Else, sorry for not understood !