GPU (CUDA CORES) FOR DSP processing?

I think the idea of using OpenCL is that the host will control access to the resources. The methods used to offload processing are transparent to the application, which IMV is nothing to really argue about because it’s a “horses for courses” situation.