scalability and flexibility of GPU clusters