When initializing a GPU context you can specify whether you want the most efficient device, the fastest device, etc, for a GUI app then you would want to pick the most efficient, which is a single flag to pass in, or supply options and let the user pick which one they want.