Calculate real power, not only rack rating
Include sustained maximum load, startup behaviour and growth reserve. Check every feed and PDU against breakers, cable size, sockets and UPS capacity. Dual power supplies provide resilience only when they connect to independent branches and either branch can carry the load after a failure.
Match cooling to the electrical load
Nearly all consumed power becomes heat. Measure temperature at server air inlets, verify hot- and cold-air separation, blanking panels and fan direction, and check for local hot spots. Dense GPU platforms may require in-row or liquid-cooling options.
Verify physical access and service space
Check usable rack depth, rail position, equipment mass, floor load and the entire delivery route. Leave enough rear clearance for cables and airflow. Engineers must be able to extend rails, replace a power supply and service cable management without disconnecting neighbouring equipment.
Prepare network and monitoring before launch
Count production, management, storage, cluster and security links separately. Confirm transceivers, cable lengths, port assignments and labels in advance. At commissioning, monitor input power, inlet temperature, UPS, cooling and leak sensors, and configure alerts before handover.
Finish with a short readiness record covering rack position, A/B power, cooling, network, grounding, monitoring and emergency shutdown. It connects the hardware design to the real capabilities of the facility.