just another stupid question/idea of mine:

how complicated/useful/performant/whatever would it be to have some layer for computation: What i mean is something that hides all computational cores and then distinguishes which/how many cores to use for a certain program call.

So every graphics would be run in software renderer using opencl on the gpu for instance (or on multiple GPUs) dividing work according to the cores (high performant GPUs for 3D works and integrated ones for 2D and say CPU for C-Ray calculations)

or is the idea just stupid as programs themselves have to decide where they want to be run on and into how many threads they can divide?

hasnt there been some article recently on CUDA for kernel uses?