Recommend a GPU shape using live capacity, memory fit, user quota, project pool, queue pressure and cost/speed objectives.