Does it support gpu acceleration for the ai tasks and can that be pointed to a separate endpoint or does it need to be local? Those specs are not too outrageous.