Configure a compute cluster for machine learning. From 1 to 20 GPUs across 1 to 5 servers each with redundant power supplies and raid 10 local SSD hard drives.