UPS Buying Guide - Waveform First, Runtime Last

The spec gate: Active PFC needs a sine wave

Almost every decision on this page collapses into one question, and it is not how many minutes of runtime you get. It is whether the unit puts out a sine wave. Modern power supplies - in servers, in desktops, in most NAS units built in the last decade - use active power factor correction. An active PFC supply does not draw current smoothly across the mains cycle; it watches the incoming waveform and shapes its draw against it. Give it a waveform that is not a sine wave and the circuit that does the watching can misread what it is looking at.

CyberPower define the compatibility feature in their own glossary as a UPS characteristic that "delivers sine wave output to prevent unexpected shutdowns or component stress for connected devices with Active PFC power supplies." Read that failure mode again, because it is the one that matters: an unexpected shutdown. Not a warning light, not degraded runtime. The machine goes off. It goes off at the exact moment the UPS switched to battery, which is the exact moment you bought the UPS to survive.

This is why the table above defaults to pure sine only and puts everything else behind a labelled disclosure. It is not an attempt to hide cheaper inventory - the counts are stated in the open and one click shows every row. It is that the default view of a page about power protection should not put a unit in front of a server buyer that may drop the server the first time the mains flickers.

The stepped or simulated wave is not a defective product. It is a cheaper inverter design that outputs a coarse approximation of a sine wave, and for a router, an access point, a network switch or a simple desktop with a passive supply it is entirely adequate and costs meaningfully less. The mistake is not buying one. The mistake is buying one for the machine you actually care about, on the assumption that a UPS is a UPS.

Why a blank waveform field is the worst answer

When we seeded this category we counted something worth stating plainly. Of the in-stock units we verified, a clear majority of those that say anything at all say pure sine. The number that state a simulated or stepped wave is zero. Not a small number - zero.

That is not because the simulated-wave units have vanished from the market. It is because pure sine output is a feature worth printing on a box and a stepped approximation is a thing you leave out of the bullet list. The silence is doing work.

So we render three states rather than two, and we do not collapse the third into either of the others. A listing that states pure sine is marked pure sine. A listing that states simulated is marked simulated. A listing that says nothing is marked not stated, in amber, and it stays that way until somebody finds a datasheet. We will not fill that cell in by inference in either direction: guessing pure sine would sell somebody a shutdown, and guessing simulated would be inventing a fact about hardware we have not tested.

What to do about it as a buyer is simple and slightly tedious. Take the model number to the manufacturer's own product page - not the retail listing, not a review, the maker's specification sheet - and look for the output waveform line. It is nearly always there. If it is not there either, that absence is now a real data point rather than a gap in a retail description, and the safe reading of it is to assume the cheaper design.

Sizing, without any number we made up

The other thing people get wrong is sizing, and the trap is built into how these products are marketed. The number on the front of the box is volt-amperes. The number that determines whether your equipment fits is watts. They are not the same number and the ratio between them is not fixed across makers.

Synology's own guidance says this outright, and we quote it rather than paraphrase because paraphrasing is where a caveat gets lost: "Different manufacturers have different volt-ampere (VA) ratings based on the performance of their UPS. To determine the total power (W) that the UPS can support, we recommend consulting the manufacturer's product manual."

Their published method is a single division. Add up the real wattage of everything you are going to plug in, then divide by the UPS's output power factor, and the result is the output power the unit needs to be rated for. Their worked example takes a 650 W load and a unit with a 70% output power factor and lands on 928.5 W - which means a box proudly labelled 1000VA, if it happens to be rated at 600 W, does not carry that load at all.

We do not supply your load figure. There is no table on this site of what a four-bay NAS draws or what a 2U server draws, and there is not going to be one until we have fetched and logged a maker datasheet for each machine. A wattage attributed to "typical" hardware is not a citation. Your own device publishes its power consumption in its specifications, usually under an environment or general heading, and that number - not ours - is the one to divide.

The one thing our own catalogue can tell you honestly is what is on the shelf at each VA band and what real power those units claim. That is a measurement of our inventory, it is labelled as one in the table above, and it is deliberately not dressed up as a recommendation about your hardware.

The battery is a consumable and it fails quietly

A UPS is the rare purchase where the thing you bought stops working without telling you. The sealed lead-acid battery inside a typical unit has an expected service life of roughly three to five years according to APC's own guidance, and heat shortens it - a unit in a warm cupboard ages faster than the same unit in a cool rack.

When that battery reaches the end of its life, the front panel does not necessarily change. The unit still filters, still regulates, still passes mains straight through to the outlets. Everything looks normal. It has simply become an expensive surge protector, and you discover this during the outage rather than before it.

Replacement cartridges for common small-business units are widely reported around the $30-50 mark. We label that as reported rather than measured because we do not track UPS battery prices and we are not going to publish a number we have not checked as though we had. Look up the specific cartridge your model takes before you count on the cheap-repair path.

Whatever you buy, the test is worth doing on day one, and the instruction comes from a vendor rather than from us. Synology's compatibility page tells readers that the power-off function varies by brand and model and to simulate a power outage to confirm it behaves as expected. Pull the plug deliberately, while you are watching, on a day when nothing is at stake. A shutdown chain that has never been tested is a shutdown chain you are guessing about.

Tower or rackmount, and the depth question nobody asks first

Form factor looks like the easy decision and is where the returns come from. A tower unit sits on the floor, takes a standard wall outlet, and does not care about your rack. A rackmount unit occupies 1U or 2U, bolts in, and puts the weight where the rest of the equipment already is - which matters, because these units are heavy and the weight is nearly all battery.

The trap on the rackmount side is depth rather than height. Racks vary, and a great many home and small-office racks are shallower than the full-depth cabinets these units were designed for. A 2U unit that is 26 inches deep does not fit a rack with 20 inches of usable depth, and the U count in the product title tells you nothing about that. Where a listing states its depth we render it; where it does not, it does not, and measuring your rack before ordering is not optional.

The second rackmount consideration is the plug. Larger units frequently ship with a NEMA 5-20P plug rather than the ordinary 5-15P, which is a physically different plug with a rotated blade and will not go into a standard household outlet. Where a listing states its plug type we show it as that listing's statement of what it ships with. What we do not do anywhere on this site is tell you what circuit to install. That is an electrician's decision and a code question, and the gap between "this unit ships with a 5-20P plug" and "go and put in a 20-amp circuit" is the gap between a fact and a liability.

What a UPS is for, which is not what most people think

It is worth being blunt about the job, because the marketing has trained everybody to shop for the wrong metric. Runtime is the headline number and it is close to irrelevant for the machine most readers of this site are protecting.

A file server does not need to keep serving files through a two-hour outage. It needs enough time to notice the power has gone, finish what it is writing, flush its caches, unmount its volumes and shut down cleanly. That is a job measured in a couple of minutes, not hours. What kills an array is not the outage; it is being switched off mid-write, repeatedly, until something in the metadata does not survive it.

Which means the signalling matters more than the battery. A UPS with a big battery and no data cable to the machine is a delay, not a protection: it postpones the ungraceful shutdown by twenty minutes and then delivers it anyway. A UPS with a modest battery and a USB cable that the operating system actually recognises will save the array. The comms column in the table above is there because it is the column that does the work.

That is also why we publish no runtime figures of our own. Runtime is a function of load, battery age and temperature, and we measure none of the three. Where a seller states a runtime we render it as that seller's claim attached to that seller's listing, and we never aggregate those claims into a number that looks like ours.

Where to go next

UPS for a NAS covers the shutdown integration in DSM, QTS and TrueNAS and what Synology’s tested list does and does not tell you · UPS for a server rack is the rackmount case, including the depth trap · tower vs rackmount compares the two on price and on what each actually costs you · 600-900VA units · 1000-1500VA units · 1500VA and up units · NAS devices and used servers are the machines on the other end of the cable · enclosures and 10GbE gear are the rest of what shares the outlet.

Frequently asked questions

Sources

Specification claims on this page were checked against the pages below on 4 August 2026. Where a widely-repeated figure did not survive that check, we report what the source says now and say so in the text rather than repeating the familiar number.

1. CyberPower Systems - Active PFC Compatibility (glossary) (cyberpowersystems.com, 2026-08-20)
   Used for: the maker's own definition, quoted verbatim on our pages: Active PFC compatibility is "a feature of UPS systems that delivers sine wave output to prevent unexpected shutdowns or component stress for connected devices with Active PFC power supplies." This is the source for the spec-gate framing - the failure mode named by the maker is an unexpected shutdown, which is precisely the event a UPS was bought to prevent.
2. Synology Inc. - How do I choose a suitable UPS? (Knowledge Center) (kb.synology.com, 2026-08-20)
   Used for: the sizing method used on these pages, quoted rather than paraphrased: "Output power = Total wattage of your devices / Output power factor of a UPS", with Synology's own worked example dividing a 650 W load by a 70% power factor to reach 928.5 W. Also the VA-versus-watts caveat in the vendor's own words: "Different manufacturers have different volt-ampere (VA) ratings based on the performance of their UPS. To determine the total power (W) that the UPS can support, we recommend consulting the manufacturer's product manual." Article last updated 1 April 2024.
3. Synology Inc. - Products Compatibility List - UPS, filtered to tested by Synology (synology.com, 2026-08-20)
   Used for: the full tested list, captured in a rendered browser session: 79 entries, 73 UPS units and 6 management cards or protocol adapters. Also the vendor's own two operational notes, which we quote rather than restate: that any listed model can extend UPS service to a second Synology product over the network, and that "The UPS power-off function varies by brand and model. Please simulate a power outage to ensure the function performs as expected." The list is CATEGORY-level: it does not filter per machine.
4. QNAP Systems - Configuring the UPS settings, QTS 5.2.x (docs.qnap.com, 2026-08-20)
   Used for: that QTS supports three connection modes - USB cable, SNMP over the network, and network UPS slave against another machine acting as the server - and the two configurable responses to a power failure: powering off after a set period, or entering auto-protection mode, which stops services and unmounts volumes.
5. Network UPS Tools project - Network UPS Tools (NUT) - project overview and hardware compatibility list (networkupstools.org, 2026-08-20)
   Used for: the project's own description of its scope - "to provide support for Power Devices, such as Uninterruptible Power Supplies, Power Distribution Units, Automatic Transfer Switches, Power Supply Units and Solar Controllers" - its GPL licence, and the size of its hardware compatibility list at the date read: at least 199 manufacturers and 1,436 device models. We link this list; we do not reproduce it. It is maintained upstream, it changes, and a stale copy of somebody else's compatibility list is worse than a link to the live one.
6. Schneider Electric (APC) - What is the expected life of my APC UPS VRLA battery? (se.com, 2026-08-20)
   Used for: the three-to-five-year expected service life of the sealed lead-acid battery in a typical APC UPS, and that operating temperature is the dominant factor in shortening it. The replacement-cartridge prices quoted on our pages are described as reported figures, not as our measurement - we do not track UPS battery prices.
7. DatacenterDisk - Live catalogue - UPS listings (datacenterdisk.com, 2026-08-20)
   Used for: every price, VA rating, watt rating, outlet count and waveform statement on these pages, read from the seller's own title and feature bullets at seed time and refreshed by the price cron. The waveform coverage rate is recomputed on each render rather than frozen here.