Currently we create too many GPU commands. We can implement some optimizations to pare this back. This is dependent on #123.
