Redundant power supply

A redundant power supply arrangement puts two or more units in one chassis. If one fails, loses AC input, or gets pulled for service, the others carry the load without a reboot. Server supplies built this way are hot-swappable, so a technician removes a dead unit and seats a replacement without touching the workload. Each unit reports status, temperature, and load to the BMC over PMBus.

The usual layout in a two-supply chassis is 1+1: a server drawing 800 W runs on two 1,200 W units, either one able to carry it alone. Larger platforms move to N+1, say four 2,700 W supplies where three cover a 6 kW load. Denser ones go to 2N, with the units split across separate A and B circuits. That split matters. Two supplies on one breaker protect against a dead unit and nothing else.

Redundancy costs parts and a little efficiency, since each supply runs at a lower share of its rating. It earns that back in production and is easy to skip on a lab machine you can afford to lose for an hour. Size total wattage to GPU count with headroom, not to the CPU and drives.

Sources

Last verified August 29, 2026.