An emergency-management InfiniBand rack is onboarded for civic AI. What should initial configuration include?

InfiniBand initial provisioning for an emergency-management civic AI rack must establish a known-good switch firmware and MLNX-OS (or equivalent InfiniBand switch OS) baseline before the fabric carries production RDMA. Firmware and OS baselines define supported features, security fixes, SM behavior, and port bring-up expectations. Embedding that baseline in the municipal onboarding checklist prevents a half-configured rack from entering the city’s emergency AI path with mismatched images or unknown switch state. Deploying Cumulus Linux baseline on InfiniBand switches fails because Cumulus is an Ethernet network operating system, not the InfiniBand switch OS; the civic fabric needs MLNX-OS (or equivalent IB switch OS) firmware and baseline, not a Cumulus image, on its IB switches. Activating OpenSM on leaf switches for distributed civic fabric discovery fails because subnet-manager activation is a fabric bring-up step, not a substitute for the required switch firmware and MLNX-OS baseline that must be in place before the IB fabric is initialized. Assigning Ethernet IP addresses to HCAs during InfiniBand provisioning fails because HCAs attach to the IB fabric using GUIDs and InfiniBand addressing, not Ethernet IP addresses, so this step confuses fabric connectivity and skips the required switch firmware/OS baseline. Exam caveat: NCP-AIN treats InfiniBand switch firmware/OS baseline as part of initial configuration, distinct from Cumulus-on-IB habits, OpenSM-as-baseline assumptions, and Ethernet IP assignment to HCAs. Operational check: after imaging, record firmware and MLNX-OS versions on every IB switch, confirm they match the civic golden baseline, verify management reachability, then attach HCAs and confirm IB ports progress to expected Up states before emergency workloads run.