operational efficiency gains with GPU clusters