Server Power Supplies: Redundancy, Sizing and Replacement

Power supplies are the component most likely to fail in a server and the one least often planned for. They contain moving parts, run continuously, and handle the electrical stress of everything else in the chassis.

This guide covers redundancy, sizing and the specifications that decide whether a replacement will actually work.

Redundancy is the point

A server with a single power supply stops when that supply fails. A server with two continues.

The common arrangement is N+1: one more supply than the load requires. With two supplies each capable of carrying the whole server, either can fail and the machine keeps running.

Two things make this genuinely useful rather than theoretical.

Hot-swap. Redundant supplies are almost always replaceable while the server runs, so a failure becomes a swap rather than an outage.

Separate feeds. The redundancy only covers supply failure if both supplies share a power source. Connecting them to separate circuits, separate PDUs or separate UPS units extends the protection to upstream failures — which is where most real outages come from.

A server with dual supplies plugged into the same power strip is protected against one specific failure and nothing else.

Sizing: bigger is not free

The wattage rating states the maximum the supply can deliver, and the instinct to buy generously is understandable and slightly wrong.

Power supplies are most efficient at a middle portion of their rated load and less efficient at very low load. A heavily over-specified supply running at a small fraction of its rating wastes energy as heat, continuously, for the life of the server.

Under-sizing is worse. A server that draws more than its supply can deliver will shut down under load, and this appears as intermittent unexplained reboots that are difficult to diagnose because they only happen when the system is busy.

What actually drives the requirement: processor TDP, the number of memory modules, drive count, and expansion cards. GPUs and accelerators dominate where present — they can exceed everything else in the chassis combined.

Most vendors publish a power configurator for their platforms. Use it rather than estimating, and size for the configuration you intend to grow into rather than the one you are ordering.

Efficiency ratings

Efficiency describes how much drawn power reaches the components rather than becoming heat. Certification tiers exist under recognisable names, with higher tiers indicating better efficiency at typical loads.

Two reasons it matters beyond the electricity bill.

Cooling. Wasted power becomes heat inside the chassis, which the fans then remove — using more power. In a rack, inefficiency compounds.

Circuit capacity. In a full rack, the limit is often the circuit rather than the space. More efficient supplies draw less for the same output, which can be the difference between fitting another server and not.

For a single server the difference is modest. Across a rack over several years it is not.

Replacements are model-specific

This is where orders go wrong, and the pattern will be familiar from drive carriers.

Server power supplies are not generic. They are built to fit a specific chassis, with a specific connector and form factor, and the server expects to communicate with them.

Physical form factor differs between platforms and often between generations of the same platform.

Connector and pinout must match the chassis backplane.

Management communication matters: enterprise supplies report status, wattage and health to the management controller. A supply the server cannot interrogate may be flagged as unsupported, or the redundancy status may not report correctly.

Wattage matching. Many platforms require redundant supplies to be the same wattage. Mixing can disable redundancy entirely, which defeats the purpose.

The reliable approach is the one that applies to every OEM part: read the part number off the supply currently fitted and order against it.

Input voltage

Worth checking because it affects available output.

Many enterprise supplies deliver their full rated wattage only on higher input voltage, and less on standard lower-voltage supply. A supply rated at a headline figure may deliver noticeably less in an office environment than in a data centre.

If you are moving equipment between facilities, or specifying for an office rather than a data hall, confirm the output at the voltage you actually have. A server that worked in a colocation rack can become power-constrained in a comms room.

Failure signs

Power supplies usually give warning, and the management controller reports most of it.

Status alerts from iLO, iDRAC or equivalent. These are the earliest indication and the easiest to act on — see our remote management guide.

Fan noise changes. Supplies have their own fans, and bearing wear is often audible before failure.

Intermittent reboots under load. Classic symptom of a supply that can no longer deliver its rated output, and easily misdiagnosed as a software or memory fault.

Redundancy lost warnings. Meaning one supply has failed and the server is running unprotected. This should be treated as urgent rather than logged and forgotten, because the redundancy that was protecting you is gone.

Holding spares

Power supplies are strong candidates for a spares shelf. They fail more often than most components, they are hot-swappable so replacement takes minutes, and they are model-specific so the market cannot be relied on at short notice.

On end-of-life platforms this matters more, since the manufacturer no longer supplies them. Our spares strategy guide covers building a position around discontinued parts.

Send us the part number from a fitted supply and we will match it. The bulk quote page explains what to include.

Common questions

Send us the part number from a fitted supply and we will match it before quoting.