A GPU server has insufficient PCIe connectivity for its intended adapters. What is the key concern?

1. The model output labels alone
2. Topology and host I/O bandwidth can constrain performance
3. The number of approved end-user accounts
4. Only the storage retention period