scalability and future-proofing of GPU clusters