UPS types: online, line-interactive, and standby for data centers

What a UPS does, and the gap it fills

A UPS, an uninterruptible power supply, is the equipment that keeps clean power on the critical load through a power disturbance and bridges the gap until the generator picks up. It does two jobs at once. It rides through the disturbance, the sag, the swell, the spike, the outright loss, from stored energy so the load never sees it. And it conditions the power, cleaning up what the utility delivers so the load runs on a steady sine wave instead of whatever shows up at the service.

The UPS is the piece between the utility and the load. The generator behind it carries a long outage, but a generator takes seconds to crank, build voltage, and accept load, and seconds is a lifetime to a server. The UPS covers that gap on battery or flywheel, then hands off to the generator. How many UPS modules you install, fed from how many paths, so the load survives a failure or a maintenance window, is the redundancy question, and it is worked in the UPS topology and redundancy design guide. This guide is about the other half: what kind of UPS, the technology inside the box.

The type decides what the load actually sees during a disturbance. Some types let the load ride on raw utility until something goes wrong, then switch. One type never lets the load touch the utility at all. That difference, whether there is a transfer and how long it takes, is the whole reason data centers settled on one type and home offices on another.

What are the three UPS topologies?

The three UPS topologies are standby (offline), line-interactive, and double-conversion online, and IEC 62040-3 is the standard that classifies them. It does it with a three-letter code that describes how independent the output is from the input. VFD, voltage and frequency dependent, is the standby UPS: the output follows the input until the input fails. VI, voltage independent, is line-interactive: the output voltage is held steady against input swings while the frequency still follows. VFI, voltage and frequency independent, is double-conversion online: the output is fully decoupled from the input, clean regardless of what the utility does.

The codes are worth knowing because a spec sheet will list one, and VFI is the class a data center is after. The progression from VFD to VI to VFI is a progression in how much the load is protected and isolated, and it tracks cost and complexity the same way. A standby unit is cheap and barely involved until it has to act. A double-conversion unit is always working, always conditioning, and costs and consumes more for it.

The table below is the short version. The sections after it work each topology in turn: how it carries the load normally, what happens when the utility drops, and where it belongs.

Standby (offline) UPS

A standby UPS, also called offline, runs the load straight off the utility and only switches to its inverter and battery when the utility fails. In normal operation the load is fed from raw mains through a transfer switch, the battery sits charged and idle, and the inverter is off. When the input drops out or drifts out of bounds, the unit transfers the load to the inverter, which builds AC from the battery. That transfer is not instant. There is a brief break, commonly a handful of milliseconds, while the switch moves and the inverter starts carrying the load.

That break is the whole story of the standby type. Most desktop power supplies tolerate a few milliseconds of gap without dropping, so a standby UPS is fine for a PC, a point-of-sale terminal, or a small piece of electronics where a short interruption is survivable. It is cheap, simple, and efficient, because nothing is converting power until it has to.

It is also the wrong tool for a data center, and not by a little. The load runs on unconditioned utility most of its life, so sags, surges, and noise pass straight through. And the transfer break, harmless to a PC, is exactly what a hall of servers cannot take. Standby UPS is for the desk, not the white space.

Line-interactive UPS and the voltage regulator

A line-interactive UPS adds voltage regulation to the standby idea, so it corrects sags and swells without going to battery. The piece that does it is an autotransformer with taps, an automatic voltage regulator (AVR), sometimes called a buck-boost. When the incoming voltage drifts low the AVR boosts it back up, and when it runs high the AVR bucks it down, all by switching taps, with the battery untouched. The load only goes to battery when the input fails outright or drifts past what the taps can correct.

This matters because brownouts and chronic low or high voltage are far more common than total outages, and on a standby UPS every one of those would drain the battery. The line-interactive unit handles them with the AVR and saves the battery for a real outage. That is why it suits small servers, network switches, telecom cabinets, and edge gear where the power is dirty but a few milliseconds of transfer on a true failure is acceptable.

It is still a transfer topology. When the utility actually drops, the load transfers to the inverter with a brief break, the same as standby, just less often. And line-interactive tops out at relatively small ratings. You will see it up to a few kVA, in racks and closets, not carrying a data hall. For that you need the type that never transfers at all.

What is a double-conversion online UPS?

A double-conversion online UPS rebuilds the power twice, so the load always runs off the inverter and never touches the utility directly. The name comes from the two conversions. A rectifier turns incoming AC into DC, feeding a DC bus. An inverter turns that DC back into clean AC, and the load runs on the inverter's output continuously, every second, whether the utility is healthy or gone. The battery sits on that same DC bus, charged through the rectifier.

The payoff is in what happens when the utility fails: nothing the load can see. The inverter is already drawing from the DC bus, and the instant the rectifier loses its source, the battery feeds the DC bus instead. There is no switch to throw and no gap to ride, so the transfer time is zero. The load was on the inverter before the failure and stays on the inverter through it.

The four pieces to know are the rectifier, the inverter, the battery on the DC bus, and the static bypass that stands ready if the inverter cannot carry the load. Because the load is always isolated behind the rectifier and inverter, a sag, a swell, harmonic noise, a frequency wander, none of it reaches the load. The utility could be a mess and the output is still a clean, regulated sine wave. That full isolation and that zero transfer are why this is the type that runs critical power.

What UPS do data centers use?

Data centers use double-conversion online UPS almost without exception, for three reasons that all come back to the load never touching the utility. The first is zero transfer time. A hall of servers cannot take even the few-millisecond break that a standby or line-interactive unit has on a real outage, and double-conversion has no break at all. The second is complete isolation. The load runs behind the rectifier and inverter, so every utility disturbance, the sag from a motor starting upstream, the swell, the harmonic distortion, the frequency drift on a generator, is absorbed before it reaches the IT gear. The third is clean output: a tightly regulated sine wave at the right voltage and frequency no matter what arrives at the input.

The simpler types fail one or more of those tests. Standby leaves the load on raw utility and breaks on transfer. Line-interactive conditions voltage but still breaks on a true outage and does not scale to data-hall ratings. Neither isolates the load the way servers need.

The cost is efficiency, since converting power twice all day long is not free, and that is the tradeoff the eco-mode and efficiency sections take up. For the critical load it is a tradeoff the industry made decades ago and has not reversed: the protection is the product, and double-conversion online is how you get it. How many of these units, and how they are arranged for redundancy, is the subject of the UPS topology and redundancy design guide.

The static bypass and the maintenance bypass

Every data-center UPS has a bypass path, the way the load keeps power when the UPS itself cannot supply it, and there are two kinds. The static bypass is internal and automatic. It is a solid-state switch that moves the load from the inverter to raw utility in a fraction of a cycle when the inverter cannot hold the load, on an overload, an internal fault, or a condition the unit cannot ride. The load stays up, but while it is on static bypass it is sitting on unconditioned utility with none of the isolation the UPS normally provides. That is the safety path. It keeps the load powered, but it is not protection, and an overload that throws the unit to bypass had better not find a fault on the bypass source.

The maintenance bypass, or wrap-around bypass, is external and manual. It is an interlocked set of switches that feeds the load from utility around the UPS so the whole unit can be powered down and serviced while the load stays up. Its rule is make-before-break: establish the bypass before the UPS is opened, so power never lapses during the switch. A UPS bought without a maintenance bypass is a UPS you cannot service without dropping the load, which is a problem you discover at the worst time.

For choosing a type, the point is that both bypasses live on the online unit and shape how it is operated and serviced. How a bypass interacts with redundancy, and how its transfers get proven during commissioning, are worked in the UPS topology and redundancy design guide.

What is eco mode on a UPS?

Eco mode is a high-efficiency operating mode where the UPS normally feeds the load through its bypass path, close to straight utility, and switches to full double-conversion only when the power goes bad. Running on bypass, the unit skips most of the conversion losses, so manufacturers cite bypass-path efficiency in the high 90s percent against the lower-to-mid 90s for full double-conversion. On a large UPS that difference is real energy and real money over a year.

The catch is exactly what double-conversion was bought to prevent. In plain eco mode the load is riding on utility, exposed to the disturbances the online unit normally absorbs, and when the utility drops the unit has to detect it and transfer back to double-conversion, which reintroduces a small transfer. That is the efficiency-versus-protection tradeoff in one line: you get the efficiency by giving up the always-on conditioning.

Modern units soften this with multi-mode or advanced eco designs, often branded eConversion or similar, where the inverter stays active in parallel and ready, so the response to a utility failure is faster than plain eco and some conditioning stays in place, at a small efficiency cost against pure bypass. Whether to run eco mode in a data center is a live debate. Many operators leave critical halls in full double-conversion and reserve eco mode for less critical loads, on the view that the conditioning is why the UPS is there at all. Confirm the mode against the load's tolerance and the manufacturer's guidance before you commit it.

Double-conversion efficiency and the losses you pay for

Double-conversion costs efficiency because the power is converted twice, all day, whether or not the utility is misbehaving, and each conversion turns a slice of the power into heat. Manufacturers commonly rate modern double-conversion units in the mid-90s percent at a good load point, with the exact figure depending on the unit, the load fraction, and the mode, so take the number from the manufacturer's curve, not a rule of thumb. The losses show up twice on the bill: once as the power the UPS consumes, and again as the cooling that has to remove the heat it made.

That second cost ties the UPS into the facility's overall efficiency, the PUE, where every watt the UPS wastes is a watt the plant pays to deliver and then to cool. It is also why partly loaded UPS efficiency matters: a unit running at a small fraction of its rating is less efficient than one well loaded, so a far-oversized UPS bleeds efficiency for years. The energy side of UPS selection is its own analysis with its own benchmarks; carry the rule that the efficiency curve, not the headline number, is what you size and operate against.

The modern lever on this is the front end and the transformer. Older transformer-based units carried more loss; today's transformerless designs run higher efficiency in a smaller frame, which is the next section.

Transformer-based versus transformerless UPS

Transformer-based and transformerless describe whether the UPS has a large internal isolation transformer, and the field has moved hard toward transformerless for size, weight, and efficiency. The older design puts an isolation transformer at the output, which gives galvanic isolation and strong fault and overload capability but adds bulk, weight, and loss. The transformerless design uses IGBT power electronics to do the job without that transformer, so the unit is smaller, lighter, and more efficient, and it fits where floor space and load capacity are tight.

The practical split runs by size and by what the design needs. Transformerless leads the small and mid range and a great deal of the large range now, on efficiency and footprint. Transformer-based units still appear at large ratings and where a design wants the galvanic isolation, the higher fault tolerance, or an output configuration the transformer provides, and some specs call for an isolation transformer for grounding or noise reasons regardless. Manufacturers and engineers put the crossover at different points, often in the low hundreds of kW, so treat it as a design call against the project, not a fixed line.

If you need galvanic isolation for the grounding scheme or the load, you either buy a transformer-based unit or add an isolation transformer downstream, and that decision belongs in the basis of design, not in a default.

The battery and the runtime to the generator

The battery is what gives the UPS its ride-through, and in a data center it is sized to bridge minutes to the generator, not to run the building. When the utility fails, the battery carries the load on the inverter while the generator cranks, builds voltage, and accepts the block, then the generator takes over and the UPS recharges. Autonomy is usually short on purpose, often a few minutes up to around fifteen, because a generator that starts in seconds does not need an hour of battery behind it. It needs enough to cover the start with margin for a start that does not go cleanly.

The chemistry choices, VRLA, flooded, and lithium-ion, and how you test and maintain them, are a full discipline of their own, worked in the UPS battery maintenance and testing guide. The short version for choosing a UPS: VRLA sealed lead-acid is the long-standing default, flooded is the long-life high-maintenance option, and lithium-ion is the growing choice for its size and life.

Whatever the chemistry, the runtime only counts if it has been proven at the real design load, because runtime falls off sharply as the load rises, and a number proven at light load is not the number you get in an outage. Treat the autonomy as a witnessed test result, not a brochure figure.

Lithium-ion UPS and the market shift

Lithium-ion is the clear direction of travel for UPS energy storage, and it is worth understanding even when you are choosing the UPS rather than the battery. The pull is density and life. Lithium holds several times the energy of VRLA in the same space and weight, commonly cited as a footprint reduction on the order of half to two-thirds, so the battery room shrinks or the same room holds more autonomy. It lasts longer, often ten to fifteen years against the three to five typical of VRLA, recharges faster, and tolerates a warmer room, which eases the cooling load. Industry trackers now have lithium passing lead-acid as the dominant UPS battery chemistry, and the push to high-density AI halls is accelerating it.

The tradeoffs are real and specific. Lithium costs more up front, it arrives with a battery management system that is part of the safety case rather than an accessory, and it carries fire and listing requirements that lead-acid does not. The system is generally listed to UL 9540 with the UL 9540A fire-propagation test method behind it, and NFPA 855 governs the installation where it applies.

For UPS selection the takeaway is narrow: more UPS lines now ship lithium-ready or lithium-native, and if the project is weighing footprint, life, and total cost over a long horizon, lithium usually wins on everything but first cost. The detail of testing and living with each chemistry stays in the UPS battery maintenance and testing guide, and the room-scale and grid-scale storage version sits in the battery energy storage work.

What is a flywheel or rotary UPS?

A flywheel or rotary UPS stores energy in a spinning mass instead of a chemical battery, and it rides through an outage on inertia. In a flywheel system a motor keeps a heavy rotor spinning at high speed; when the utility fails, that rotor's momentum drives a generator and carries the load for the seconds it takes the standby generator to start and pick up. Ride-through is short by design, commonly on the order of 10 to 30 seconds at full load, because a flywheel stores far less energy than a battery bank. That is enough given that most utility disturbances last only seconds and a well-maintained generator starts inside that window.

The rotary idea taken further is the diesel rotary UPS, the DRUPS, which couples the spinning mass, a generator, and a diesel engine on one shaft. The inertia covers the gap, then the diesel clutches in and the same machine that was riding through becomes the generator carrying the outage, with no separate UPS-to-generator handoff. DRUPS shows up on large, single-vendor critical-power blocks where its compactness and the absence of a battery room are worth the specialized maintenance.

The appeal of both is no battery: no cells to age, no battery room, no thermal-runaway risk, and a known, repeatable ride-through. The cost is that the energy reserve is measured in seconds, so a flywheel design leans hard on the generator starting reliably and fast. Pair a thin ride-through with a generator that cranks slowly and you have a thin bet, which is why some sites add a short battery behind the flywheel or hold the generator to a tight start spec.

Modular UPS, monolithic, and form factors

A modular UPS is a frame populated with hot-swappable power modules, and it changes how you buy and grow capacity. Instead of one fixed-rating monolithic unit, you add modules as the load grows and pull and replace a failed module without dropping the load. The redundancy can live inside the cabinet: size the frame so the load is covered with one module out and you have N+1 in a single footprint. The pull is matching capital to load, a hall that fills over years does not pay for full UPS capacity on day one, and faster repair, since a module swaps in minutes where a monolithic unit is a much longer job.

Monolithic units still have their place. They tend to lead at the largest single ratings and where a design wants one large, simple, proven block rather than many modules and the shared logic that ties them together. The choice is capacity matching and serviceability against simplicity and top-end rating.

Form factor follows the load. Small UPS, up to a few kVA, are rack-mount units that live in the rack with the gear they protect, sized in rack units. Larger UPS are floor-standing, free-standing cabinets or lineups in the electrical room, from tens of kVA into the megawatts. The rule is plain enough: rack-mount for the closet and the single rack, floor-standing for the room and the hall. Match the form factor to where the load is and how much of it there is.

How do you size a UPS? kW, kVA, and power factor

You size a UPS from the real load in kW, then add redundancy, headroom for growth, and the runtime to the generator, and you check the rating in both kW and kVA. Start with the actual measured or calculated load, not the sum of nameplates, because nameplate totals run high and oversizing bleeds efficiency. Common practice is to load a UPS to around 80 percent of its rating, leaving room for peaks, for growth, and for installing a replacement alongside the old unit before retiring it.

The kW-versus-kVA distinction is where sizing goes wrong. kVA is apparent power, the volts times amps the unit handles; kW is real power, what the load actually consumes. Power factor is the ratio, kW divided by kVA. An older UPS rated 100 kVA at 0.8 power factor delivers only 80 kW, so a load that needs 90 kW would overrun it even though 90 is below 100. Modern data-center UPS are built for unity power factor, rated so kVA and kW are the same number, because modern IT power supplies present a near-unity load. Size on the kW the load draws and confirm the unit delivers that kW at the load's power factor, not just the kVA on the label.

Then layer on what the load alone does not tell you. Redundancy decides how many modules or systems sit around the load, the N+1 or 2N question worked in the UPS topology and redundancy design guide. Runtime decides the battery: enough autonomy to bridge to the generator at the real design load, with margin for a start that does not go cleanly. And the static bypass and the upstream overcurrent protection have to be rated for the fault and overload the system will see, not the steady state. A UPS sized to the nameplate with no margin is a UPS the first load increase overruns.

UPS input, harmonics, and the front end

The UPS rectifier is a non-linear load, so what it draws on its input matters to the upstream gear and to the generator it falls back on. Older six-pulse rectifier front ends pulled heavily distorted current, putting harmonics back onto the upstream system that distorted the voltage and forced oversized transformers and generators to absorb them. The harmonic current also loaded neutrals and could disturb other equipment on the same source.

Modern data-center UPS use active or low-harmonic front ends, IGBT rectifiers that draw near-sinusoidal current at high, often near-unity, input power factor. That cuts the harmonic footprint sharply and eases the upstream sizing, but it does not erase the question. The input filter and the front-end type still get confirmed against the generator and the upstream equipment, because the interaction is where trouble hides.

The generator case is the one to watch. A generator is a softer, higher-impedance source than the utility, so a UPS that behaves on utility can misbehave on generator: input filters can interact with the generator's voltage regulation, and the UPS recharging its battery the moment it lands on the generator is a step load that swings the generator. The fix is matching the UPS input and its battery-recharge behavior to the generator, often with a walk-in or soft-start charge, and sizing the generator for the UPS input including that recharge step. A UPS that works on utility is not proven until it works on generator, because the generator is the source you fall back to when utility is the thing that failed.

The data-center choice

For a data center the type decision is largely settled, and the live decisions are underneath it. The type is double-conversion online, because the load cannot take a transfer and needs the isolation, and the simpler topologies do not clear that bar. That much is not really a choice for critical IT load. What is a choice is everything below it: transformerless over transformer-based for efficiency and footprint unless the design needs the isolation, modular over monolithic where capacity matching and fast repair matter, and lithium-ion over VRLA where footprint, life, and long-term cost outweigh first cost.

The eco-mode question stays open and load-specific. Run full double-conversion where the load needs the conditioning, which for most critical halls is everywhere, and reserve the higher-efficiency modes for loads that can tolerate them, on the manufacturer's guidance and the load's real tolerance.

Two things belong to the redundancy side, not the type side, and the UPS topology and redundancy design guide carries them: how many of these units you install and how they are arranged, N+1, 2N, or beyond, and how the bypass and transfers prove out in commissioning. Pick the type for the load, then hand the count and the arrangement to the redundancy design. The type protects the load from the power. The redundancy protects the load from the UPS.

What to document

The UPS selection is defensible only if the record says what was chosen and why, so the next engineer and the operations team can see the basis. A one-line that shows a UPS but never states its type, mode, rating basis, and runtime leaves everyone guessing, and guessing is how an eco mode gets left running on a load that needed conditioning.

Record the type and IEC class, the operating mode the unit is set to, the kW and kVA rating with the power factor it was sized at, the transformer configuration, the energy store and its proven runtime at design load, and the redundancy the unit is part of. Tie each entry to the basis of design so a later load increase or a mode change gets checked against the intent, not against habit.

Common mistakes

• Specifying a line-interactive or standby UPS for a data hall, where the transfer break and the lack of isolation are exactly what the load cannot take.
• Running eco mode on a critical load that needs full conditioning, trading the protection the UPS was bought for to save a few percent of energy.
• Buying a UPS with no maintenance bypass, so the whole unit cannot be serviced without dropping the load.
• Sizing the unit to the kVA on the label and overrunning its kW once the load's power factor is taken into account.
• Sizing the battery or flywheel runtime short of the real worst-case generator start and load-acceptance time.
• Not understanding the static bypass, so an overload throws the load onto unconditioned utility nobody planned for.
• Oversizing the UPS so far that it runs at a low load fraction and bleeds efficiency for years.
• Proving the UPS on utility and never proving it feeds and recharges cleanly on the generator.

Field checklist

Standards and references

IEC 62040, the international standard for UPS, is the framework for the topologies and their performance, with IEC 62040-3 giving the VFD, VI, and VFI classification used through this guide along with the test methods behind the ratings. In North America the UL 1778 standard covers UPS safety, and the equipment carries the listing the jurisdiction recognizes. Name the standard that governs the point and confirm the current designation and edition, because they shift over time.

For efficiency, the ENERGY STAR program and the relevant regional efficiency rules set test methods and thresholds for UPS efficiency, and manufacturers publish efficiency curves against them. Treat a published efficiency as the manufacturer's figure at a stated load and mode, not a guarantee at your operating point. For lithium-ion energy storage, the system is generally listed to UL 9540 with the UL 9540A fire-propagation test method, and NFPA 855 governs the installation where it applies, worked in the battery and storage guides. IEEE guidance covers emergency and standby power and the powering and grounding of sensitive electronic equipment.

Above all of these sit the manufacturer's documentation and the project basis of design. The manufacturer's data governs the specific unit, its ratings, its modes, its bypass behavior, and its generator compatibility. The basis of design sets the type, the redundancy, and the runtime the project has to meet, and where it is stricter than a standard it controls. Confirm every figure against them before committing a selection.

Units, terms, and synonyms

UPS terminology overlaps and the same idea travels under several names across a spec, a manufacturer sheet, and a one-line, so a short glossary keeps the selection conversation honest.

The topologies go by pairs of names. Standby is also offline and IEC class VFD. Line-interactive is IEC class VI. Double-conversion online is IEC class VFI. Capacity is rated in kVA, apparent power, and kW, real power, related by the power factor. Runtime is the autonomy, the minutes the energy store carries the load before the generator does. Eco mode and high-efficiency mode name the bypass-favoring operating mode, and eConversion and multi-mode name the advanced versions.

Standby / offline / VFD

Load runs on utility, transfers to the inverter on failure with a brief break

Line-interactive / VI

Regulates voltage with an AVR, still transfers to battery on a true outage

Double-conversion online / VFI

Load always on the inverter, zero transfer, full isolation from the utility

Static bypass

Automatic solid-state switch to raw utility on UPS overload or fault

Maintenance bypass

Manual interlocked path to service the whole UPS with the load up

Eco mode

High-efficiency mode feeding the load through the bypass path

kVA / kW / power factor

Apparent power, real power, and their ratio; modern UPS rate at unity

DRUPS

Diesel rotary UPS, riding through on a spinning mass coupled to a diesel engine

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What does N+1 redundancy mean?
• How do you test a UPS battery?
• What is an as-built drawing?
• Why do battery rooms need ventilation?
• What is the maximum cable tray fill?