UPS Sizing and Power Protection for Servers

A UPS is bought to keep servers running through a power cut, and that is the least important thing it does. The larger value is protecting equipment from the electrical events that happen constantly and shorten hardware life quietly.

What a UPS actually protects against

Complete power loss is the visible case. Four others are more frequent and less noticed.

Sags and brownouts — voltage dropping below normal, often when heavy equipment starts nearby. Systems may keep running while components are stressed.

Surges and spikes — brief over-voltage from switching events or weather. These degrade power supplies over time rather than destroying them instantly, which is why the damage is attributed to something else.

Frequency variation — relevant on generator supply, where output is less stable than mains.

Very brief interruptions — too short to notice as an outage, long enough to reset equipment. These cause the unexplained reboots people spend days investigating in software.

Power supplies handle a certain amount of this, and repeated stress is cumulative. A UPS between the supply and the equipment removes it.

The three topologies

The specification that most affects both protection and price.

Standby passes mains through normally and switches to battery when it detects a problem. Cheapest, with a brief transfer delay, and it provides limited conditioning. Suitable for individual workstations rather than servers.

Line-interactive adds automatic voltage regulation, correcting sags and over-voltage without using the battery. This is the sensible minimum for servers and network equipment, and it is where most small installations should sit.

Online double-conversion continuously converts incoming power to DC and back to AC, so equipment always runs on regenerated power. There is no transfer time at all and the output is isolated from every input disturbance. It is the most expensive, and it is what critical systems and generator-backed sites use.

The practical guidance: line-interactive for most server rooms, online for anything genuinely critical or on unstable supply.

Sizing: two separate numbers

UPS capacity is quoted in VA and in watts, and they are not interchangeable.

Watts is real power — what the equipment actually consumes.

VA is apparent power, which is higher, and the ratio between them varies.

A UPS has limits on both, and you can exceed one without exceeding the other. Size against the watt figure using your actual load, then check the VA rating covers it too.

Measure the real load rather than adding up power supply ratings. A server with dual power supplies rated at some wattage does not draw that continuously — the rating is a maximum, and actual draw is typically well below it. Sizing from nameplate ratings produces a UPS two or three times larger than needed. Our power supply guide covers what actually drives consumption.

Most enterprise servers report actual power draw through the management controller, which is the number to use.

Runtime: what it is for

The common mistake is buying runtime as though the UPS should carry the load through an outage.

For most installations that is not the goal. The goal is enough time to shut down cleanly, which is typically minutes rather than hours.

What that requires: enough runtime for the operating systems to complete an orderly shutdown, plus margin, plus the ability to ride through brief interruptions without shutting down at all.

Extending runtime to hours means substantially more battery, at substantial cost, and it still ends eventually. Where genuine long-duration operation is needed, that is a generator with a UPS bridging the start-up gap — a different design.

Also worth planning: runtime falls as load rises, and it falls as batteries age. A UPS sized for exactly the runtime you need at installation will not deliver it in three years.

Shutdown integration is the part that gets skipped

A UPS with no shutdown integration provides runtime and nothing else. When the battery is exhausted, everything loses power abruptly — which is the outcome you bought the UPS to avoid.

Proper integration means the UPS communicates with the servers, and the servers shut down cleanly on a defined trigger, in a defined order.

Three things to get right. Connectivity — network cards for the UPS are usually a separate purchase and worth budgeting. Shutdown order — virtual machines before hosts, applications before databases. And testing, because an untested shutdown sequence is an assumption, not a plan.

This is the same principle as backup: the equipment is not the protection, the tested procedure is. See our backup guide.

Batteries are consumable

The part most often forgotten until it matters.

UPS batteries degrade whether used or not, and a unit that has run for years without a power cut may have batteries that will not carry the load for more than a moment.

Two consequences. Test under load periodically rather than trusting a status light. Many UPS units self-test and report, but a real load test is the meaningful check. And budget for replacement as a recurring cost rather than a surprise.

Heat accelerates battery ageing considerably, which connects to rack conditions — a UPS in a hot comms room loses battery life faster. See our airflow guide.

What to protect

Servers are obvious. Two things frequently left out.

Network equipment. A server that stays up on UPS while the switch it connects to loses power is unreachable. Protect the network path, not just the endpoints.

Storage arrays. Shared storage losing power while hosts stay up produces exactly the situation you were avoiding.

Conversely, one thing worth checking: enterprise SSDs include power loss protection that flushes cached writes on power failure. That protects the drive; it does not protect the transaction, the application state or the array. The UPS is still doing a different job.

Common questions

Tell us your actual measured load and shutdown requirements and we will help size the protection properly.