Rack Power Budget and Uplink Cabling for a 12-Node Compute Build
You are racking twelve 1U dual-socket servers into a 42U cabinet. Each server ships with two 750 W hot-plug power supplies, and the vendor sizing tool rates actual draw at 480 W per node under your workload mix. The cabinet has two 30 A single-phase 208 V rack PDUs, one on feed A and one on feed B, each derated to 80 percent (24 A). All twelve servers must be dual-corded with one PSU per feed, and two 1U 10GBASE-T top-of-rack switches also sit in the cabinet, each drawing 120 W from feed A. A colleague proposes putting all twelve servers' PSU1 cords and both switches on the A PDU, all PSU2 cords on B, and cabling every server to the ToR with 8 m Cat 6a patch cords. Cabinet intake temperature measured at the top U is 38 C, and the rack is missing several blanking panels.
1. Work out the steady-state load on each PDU in watts and amps and decide whether the colleague's power plan fits inside the derated capacity. Show the arithmetic.
Read the worked answer
Feed A carries all twelve servers plus both switches: 12 x 480 W = 5,760 W, plus 2 x 120 W = 240 W, so 6,000 W. On 208 V single phase, I = P / (V x PF); even at unity power factor that is 6,000 / 208 = 28.8 A, and at a realistic 0.95 PF it is about 30.3 A. The derated ceiling is 24 A (80 percent of 30 A), so feed A is over by roughly 5 A and, at PF below 1, brushes the 30 A breaker rating as well. Feed B carries only the servers: 5,760 W = 27.7 A at unity, about 29 A at 0.95 PF, so it also exceeds 24 A. Both feeds fail before any growth, PSU inrush, or fan ramp at high inlet temperature is counted. Do not size on the 750 W nameplate, but do keep a margin above the measured 480 W. Workable fixes: split into two cabinets (six nodes plus one ToR per side is roughly 3,000 W, about 14.4 A), drop to eight nodes per cabinet (about 18.5 A), or move to 60 A three-phase PDUs. Note the plan is also unbalanced: feed B has no switch load at all.
2. Evaluate the cord, PDU outlet, and copper cabling choices, and identify every redundancy gap in the proposal.
Read the worked answer
A 750 W PSU uses a C14 inlet and a standard C13/C14 jumper rated at 10 A, which is ample for about 2.3 A per server at 208 V, but it means the PDU must have enough C13 outlets; C19 outlets only matter for high-wattage PSUs with C20 inlets. Confirm the PDU input plug matches the facility receptacle, since a 30 A PDU usually arrives with an L6-30P and cannot be plugged into a 5-20R. On copper, Cat 6a supports 10GBASE-T to 100 m, so 8 m is electrically fine, but for in-rack links a passive SFP+ DAC is normally cheaper, cooler, and lower latency, and 10GBASE-T PHYs add roughly 2.5 to 4 W per port on both ends; verify the switch actually supports the 10GBASE-T optic before ordering. Redundancy gaps: the servers are correctly split across A and B, but each ToR is single-corded to feed A, so losing that feed takes down the whole top-of-rack layer. Give each switch an A and a B supply, and keep patch cords dressed so they cannot obstruct front-to-back airflow.
3. The 38 C reading at the top of the rack: what does it indicate, and what do you do before adding load?
Read the worked answer
Intake above roughly 27 C is outside the recommended operating envelope for most servers, and 38 C is beyond the allowable class for many models, so fans will run near maximum, PSUs derate, and the 480 W per node figure stops being representative because fan power rises. A top-of-rack reading that is much hotter than the bottom almost always means hot-air recirculation: missing blanking panels, open U space, unsealed cable cutouts, or a rack drawing exhaust from the hot aisle. Remediate before energizing anything new: install blanking panels in every empty U, seal floor and cable openings, confirm the rack faces the cold aisle and that the server bezels match the airflow direction, and take readings at the bottom, middle, and top of the front of the rack rather than one point. Check that perforated tiles in front of the rack are adequate for the kW load and that the CRAH units are not running short of capacity. If the reading stays high after containment fixes, raise it with facilities; do not simply open the doors, which shifts the problem downstream.