One rebuilds utility power from scratch, all the time, so the load never sees a switch. The other rides straight through on utility power and only steps in when it must — a difference that shows up the instant a transfer happens.
Not all UPS systems protect a load the same way, even though both are commonly sold as "UPS." A double-conversion (online) UPS rectifies incoming AC to DC and then inverts it back to AC continuously, so the load is always powered by the UPS's own inverter output, isolated from raw utility power at all times. A line-interactive UPS instead passes utility power through to the load nearly directly, using a transformer or autotransformer to trim minor voltage sags and surges in place, and only switches the load onto battery-fed inverter power during an actual outage or a voltage excursion outside its correction range. That switch takes a few milliseconds — transfer time — during which the load momentarily sees no clean power at all. For most IT equipment that transfer time is invisible. For a facility running thousands of GPUs mid-training, it is not always negligible, which is why virtually all Tier III/IV-class and hyperscale AI data center UPS plants use double-conversion topology despite its higher cost and lower efficiency.
In a double-conversion UPS, the inverter is never idle — it is the load's only power source at all times, continuously fed by a DC bus that the rectifier keeps charged from utility and the battery backs up. When utility fails, nothing has to switch: the same inverter keeps running off the same DC bus, just drawing from the battery instead of the rectifier, which is why the transfer is seamless with zero switching gap. A line-interactive UPS instead treats the inverter as a standby resource: normal operation routes utility power almost directly to the load through a transformer, and the inverter only comes online — via an actual transfer switch closing — when an outage or out-of-range voltage event occurs. That switch, however fast, is a real event with a real (if brief) gap, and it is why line-interactive designs are rated for a specific transfer time rather than "zero."
That's true for most standard IT equipment, whose power supplies have internal hold-up capacitors that ride through a few milliseconds without issue — which is exactly why line-interactive UPS units are perfectly adequate, and much more efficient, for typical office and light commercial IT loads. It is not automatically true for large-scale AI/GPU racks, where power supplies operate closer to their continuous rating and where a synchronized fleet of thousands of GPUs is more sensitive to any voltage transient hitting many nodes simultaneously. That sensitivity, combined with the outsized cost of a training run restarting from a checkpoint after even a brief glitch, is the practical reason hyperscale and Tier III/IV-rated AI data center UPS plants almost universally specify double-conversion (online) topology rather than line-interactive, despite line-interactive's efficiency and cost advantages — the efficiency gap of a few percentage points is a much smaller cost than a training run restart.
Explains the difference between double-conversion (online) UPS topology, where the inverter continuously supplies the load with zero transfer gap, and line-interactive UPS topology, where utility power passes through directly and the inverter engages via a transfer switch only during an outage.
A line-interactive UPS routes utility power through an autotransformer that can boost or buck (trim) minor voltage deviations in place, feeding the load nearly directly under normal conditions. Its battery-fed inverter sits in standby and only engages — through an actual transfer switch closing — when utility fails or drifts outside the autotransformer's correction range. That switch takes roughly 2–10 milliseconds depending on the unit, a brief but nonzero interruption. Because power isn't continuously converted, line-interactive units are typically 97–98% efficient and lower cost, making them the standard choice for most enterprise and light commercial IT loads.
A double-conversion UPS rectifies incoming AC to DC continuously, maintaining a live DC bus that is simultaneously float-charging the battery and feeding an inverter that is always the load's sole power source. The load never sees raw utility power directly, and a utility outage requires no switching at all — the same inverter simply keeps drawing from the battery instead of the rectifier. This gives a true zero-transfer-time seamless transition, along with continuous voltage/frequency regulation and isolation from utility disturbances, at the cost of continuously converting every watt twice, which typically runs 94–97% efficient and generates more waste heat that the facility cooling system must remove.
Standard server power supplies have internal hold-up capacitors that ride through a brief line-interactive transfer without issue, which is why line-interactive UPS units serve most conventional IT loads well. Large synchronized AI/GPU clusters are less forgiving of even brief transients hitting many nodes at once, and the cost of an interrupted multi-day training run — potentially losing substantial compute time back to the last checkpoint — dwarfs the modest efficiency advantage line-interactive offers. This asymmetry, combined with Tier III/IV requirements for concurrent maintainability and fault tolerance, is why virtually all hyperscale and enterprise-grade AI data center UPS plants specify double-conversion topology.
Yes — because every watt passes through both a rectifier and an inverter continuously, double-conversion units dissipate more conversion loss as heat than line-interactive units under normal (non-outage) operation, which is why they run at lower overall efficiency and add to a facility's cooling load and PUE.
For loads that can tolerate a few milliseconds of transfer time — the vast majority of standard IT and enterprise equipment — yes, and its higher efficiency reduces both energy cost and cooling load. It becomes a poorer fit specifically for large synchronized compute clusters or any load whose specification requires zero transfer time.
Yes — "online UPS" and "double-conversion UPS" are used interchangeably in the industry; both describe the same topology where the inverter continuously supplies the load and the load never sees raw utility power.
Some modern double-conversion UPS units offer a selectable eco-mode that behaves more like a line-interactive design under normal conditions — bypassing continuous double conversion to improve efficiency — while still switching to full double-conversion protection when utility quality degrades. This is a deliberate efficiency/protection tradeoff the facility engineer chooses, not a separate UPS category.