Line-Interactive vs Online UPS: Which Do Your Servers Actually Need?

Both UPS types keep servers running when utility power fails. They differ in what happens the rest of the time, and in the fraction of a second when the power actually goes. That difference decides which one belongs in your rack. The short version: line-interactive is the efficient, economical choice for small racks on decent power, and online double-conversion is the default for GPU servers, larger loads and anywhere power quality is a problem.

What is a line-interactive UPS?

In normal operation, a line-interactive UPS passes utility power through to its outlets. An autotransformer adjusts the voltage up or down when it sags or swells, a function called automatic voltage regulation (AVR), without touching the battery. When power fails or strays outside the acceptable range, a relay transfers the load to the inverter, which runs from the battery. That transfer takes a few milliseconds.

Server power supplies are designed to hold their output up for a short interval when input disappears, and a properly sized line-interactive UPS with sine wave output transfers well within that window. The APC Smart-UPS SMT family, such as the SMT1500RMI2U (1000W / 1500VA) and SMT750RMI2U (500W / 750VA), and the APC SMTL2200RMI2UC are line-interactive rack units.

What is a double-conversion UPS?

An online double-conversion UPS rectifies incoming AC to DC, which charges the battery and feeds an inverter, and the inverter produces fresh AC for the load, continuously. When utility power fails, the inverter simply keeps running from the battery. There is no transfer and no break. Because the output is regenerated rather than passed through, the load is isolated from sags, surges, noise and frequency drift on the input. That makes online the usual choice wherever input power is simply unreliable, such as some industrial, remote and edge sites, as well as in data centers.

The cost is efficiency. Converting all the power twice produces losses and heat, which is why many online units offer a high-efficiency or eco mode that bypasses the double conversion when input power is clean. Examples in our range include the Vertiv Liebert GXT5 1500VA (1350W / 1500VA), the APC SRTL3KRM1UIC (3000W, on-line), the Eaton 9PX 6000VA (5400W / 6000VA, with four C13 and four C19 outlets on a 230V hardwired input) and the APC SRTL8KRM4UI (8000W, online).

Line-interactive or online UPS for servers?

• Protection: line-interactive corrects voltage and transfers in milliseconds; online isolates the load completely and never transfers.
• Efficiency and heat: line-interactive runs more efficiently in normal operation; online gives up some efficiency unless it has an eco mode.
• Generators: online handles a generator's frequency and voltage variation without going to battery; line-interactive may transfer to battery repeatedly on unstable generator output, draining it.
• Capacity range: line-interactive rack units are common at smaller sizes; above a few kVA, and for three-phase power, most units are online double-conversion.
• Cost and noise: line-interactive is less expensive per watt and usually quieter; online units cost more and run cooling fans continuously.

Choose line-interactive for a small rack of general-purpose servers, network gear and storage in an office or edge site with stable utility power and no generator. Choose online when the equipment is critical, the load is large, power quality is poor, or a generator is in the path.

There is a third topology, standby (also called offline), found mainly in desktop units. It keeps the load on raw utility power with no voltage regulation and switches to battery only when power fails. It is fine for a single PC but is not a sensible choice for servers, because every sag or swell either passes straight through or sends the unit to battery. For server racks, the real decision is between line-interactive and online.

Do GPU servers need an online UPS?

In practice, usually yes, for three reasons. First, capacity: GPU servers draw far more than most line-interactive units deliver, and the rack UPS sizes that can carry them are mostly online designs. Second, load behavior: GPU load can change sharply as jobs start and stop, and an online UPS supplies that load from a continuously running inverter rather than passing input disturbances through. Third, consequences: a power event that resets a training node loses all progress since the last checkpoint, and multi-node jobs can lose more than that. Whichever topology you choose, size it with real headroom rather than running close to the rating, because GPU servers are exactly the load that finds the limit.

The exception is a single workstation-class GPU machine or a small inference box on clean power, where an adequately sized line-interactive unit is a reasonable choice. At the other end, a rack of GPU servers often exceeds what any single-phase rack UPS can carry, which leads to three-phase online units such as the APC Easy UPS 3S 20kVA (400V, double conversion) or to UPS-backed power supplied by the data center. Our article on power and cooling for 1000W GPU servers covers sizing at that scale.

Line-interactive fixes the power you have. Online replaces it. The more a dropped job costs, the more the second one is worth.

Does my colocation facility make a rack UPS unnecessary?

Often, yes. Many colocation and enterprise data centers deliver A and B feeds that are already backed by facility UPS and generators. Adding a rack UPS downstream mostly adds a failure point and heat. Confirm what your provider guarantees, then spend the budget on dual power supplies split across both feeds. The redundancy and failover planning guide covers that design. If you still want local protection, for example to control shutdown sequencing, put the rack UPS on one feed only, so a UPS fault cannot take down both power supplies of a server.

What else should I check before choosing?

• Output waveform on battery: servers with active PFC supplies should be on a UPS with sine wave output.
• Input voltage and plug: many rack models are 208V or 230V; US 120V models such as the APC SMT2200RM2UC exist for smaller loads.
• Outlet types: high-wattage server supplies may need C19 outlets rather than C13.
• Battery: lithium-ion models such as the APC SRTL3000RM4UXLI (2700W) are chosen for longer service life and lower weight.
• Management: a network card for shutdown signaling and monitoring is worth specifying on any server UPS.
• Bypass: online units with an internal static bypass keep the load powered if the inverter faults, and a maintenance bypass lets you service or replace the UPS without shutting servers down.
• Runtime: topology does not set runtime. Battery capacity at your actual load does, so check the runtime chart either way.

Nexus Compute supplies new line-interactive and online UPS units from APC, Eaton and Vertiv through authorized distribution, with the full manufacturer warranty. Tell us the load, the input power and whether a generator is involved, and we will recommend a topology and size and quote within 48 business hours.