For only 900, you should definitely have a global size of 900. Even that is a bit low. I'd recommend 1000-2000 at a minimum, more for larger GPUs.

As for local size, if you have no synchronization requirements, then it will be device-dependent and you'll have to play with it. To start you can just pass in NULL for the local size. If you have synchronization requirements then that will dictate what you need.