Control a core or a multiprocessor


I’m new about CUDA programming and development and maybe this is a crazy question! I would like to know if it’s possible to control a single core on a multiprocessor or a single multiprocessor using a particular API?



Right now, the answer is no to both. Both CUDA APIs only permit whole of GPU control. Preliminary information suggests that multiprocessor level API control will be available on the new Fermi hardware, but that is just based on preliminary information.