I am looking to setup a small lab serving multiple users and models off a single-node DGX platform (probably H200 or similar). I am looking to understand what hardware I need. I am brand new to this and need to understand if I need hardware beyond the bare box. I looked at the BasePod architecture, for example, and saw some nodes for Kubernetes, etc.
Can I get by with the storage on the bare box, or do I need a dedicated storage subsystem?
Do I need add’l nodes for management and orchestration?
Any insights appreciated.
P.S. I read this Deploying Rich Cluster API on DGX for Multi-User Sharing | NVIDIA Technical Blog. Is it still considered the best path forward for single-node deploys?