I have a few quick questions regarding Dual DMAs. NVIDIA published a white paper regarding Dual DMAs in their Quadro’s and I was wondering if that meant they also worked in OpenCL. The paper only mentions CUDA.
I currently am running a Quadro K5100M where I have a project that utilizes Compute and Transfer Overlap but when I profile it, the reads and writes are not overlapping (which would happen with Dual DMAs). I just wanted some clarification whether maybe I’m doing something wrong or it is just not possible in OpenCL.