How customers are isolated
Dedicated GPUs and storage per customer.
Each customer's GPU nodes and storage are theirs alone for the term. Accelerators are never shared between customers. Customers with stricter requirements can ask for separate hardware arrangements.