Picture this: a rack of networking gear drawing 2.2 kW behind a UPS rated 3000 VA / 2700 W. The Eaton 9PX brochure says "0.9 output PF". The APC Smart-UPS Online SRT at the same kVA says "Unity PF" on the 6–10 kVA models, and 0.9 on the mid-range. But the gear pulls 2.2 real watts. Which UPS actually delivers without derating? This is where datasheets stop being helpful — and where the deeper conversation begins.
1. Real Watts vs. Apparent Watts — The Power Factor Gap
The Eaton 9PX (700 VA–11 kVA) is specified at 0.9 output power factor, meaning a 3 kVA unit can deliver ~2.7 kW. Schneider UPS's APC Smart-UPS Online SRT goes further: on the 2.2–5 kVA models it also rates 0.9 PF, but on the 1–1.5 kVA and 6–10 kVA models it hits Unity PF — 1.0 — so a 10 kVA SRT delivers a full 10 kW. The mechanism here is the inverter design. Unity PF requires a larger DC bus capacitor bank and a control loop that can sustain rated current at unity phase angle under heavy nonlinear loads. That costs more, and adds about 2–3% I²R loss in the output filter at full load. The worked consequence for a B2B buyer: if you spec a 10 kVA Eaton 9PX, you backstop at 9 kW usable; the equivalent SRT gives you 10 kW — a ~11% headroom difference on the same frame size. The reversal: if your load power factor is 0.8 lagging (typical for older server PSUs with no active PFC), both units deliver the same ~80% of kVA, and the Unity PF advantage evaporates. For modern PFC rectifier loads (PF >0.95), the Unity PF unit is measurably more efficient in real watts delivered per rack U.
2. Efficiency Mode — What "98% Efficiency" Really Means
Schneider's Galaxy VS (10–150 kW three-phase) claims up to 97% in double-conversion, and eConversion high-efficiency mode up to 99%. Eaton UPS's 9PX is ENERGY STAR qualified but does not publish a specific double-conversion efficiency number — the brochure says "high-efficiency operation" without a percentage at typical load. The mechanism is the operating mode: eConversion is a Class-1 no-break transfer that keeps the inverter running synchronised to bypass, but with the rectifier off, so only the inverter's switching losses (~1–2%) remain. If the line fails, the transfer to double-conversion or battery is seamless — zero break. The worked consequence for a 50 kW data center load: running Galaxy VS at 99% eConversion vs. 97% double-conversion saves about 1 kW of heat per 50 kW load — roughly 3,400 BTU/h — which can mean one fewer cooling rack unit or a lower PUE. The reversal: eConversion only works when the input voltage and frequency are within ±5% of nominal and the load power factor is above 0.9. On a noisy generator feed with frequency swings >1 Hz, the system falls back to full double-conversion (97% eff), and the brochure's 99% number no longer applies. The datasheet hides that constraint.
3. Transfer Time — Zero vs. "Zero" Under Load Transient
Both the Eaton 9PX and Schneider APC Smart-UPS Online (SRT) are double-conversion (VFI) topologies. Transfer time is zero in theory — the inverter always feeds the load. But real transfer time appears during a bypass-to-inverter transition when the inverter synchronises. The mechanism: in double-conversion, the inverter's output is always online; a static switch transfers the load between inverter and bypass in under 2 ms if the inverter fails or is overloaded. That's a break-before-make event. APC's SRT claims "zero transfer time" — which is true for the normal inverter-to-battery transition because the battery is in the DC link. The worked consequence: for a server with a hold-up time of ~10 ms (typical for modern ATX PSUs), a 2 ms break is invisible. But for a PLC or medical device with a hold-up under 1 ms, a 2 ms gap can cause a reset. The reversal: Eaton's 9PX datasheet does not specify the static-switch transfer time — they rely on the VFI classification to imply "no break". If you need guaranteed zero break (not just <2 ms), the APC SRT's "zero transfer" is the safer claim, but only because they define it as continuous inverter operation, not static-switch bypass.
4. Runtime Under Load — The Curve They Don't Print
For the same 3000 VA class, example runtime data from Tripp Lite's SmartOnline SU3000RTXL3U (a sibling product, same topology class as Eaton's 9PX) shows ~14 min at half load (1200 W) and ~5 min at full load (2400 W) on internal batteries. CyberPower's OL1000RTXL2U (1000 VA / 900 W) gives ~15 min at half load and ~5.9 min full load. Neither Eaton nor APC publishes a runtime curve for the 9PX or SRT in their short-form datasheets — they list "runtime varies by load". The mechanism is battery capacity vs. inverter efficiency: at light load (20–30%), the inverter's fixed losses (control power, fans, magnetics) consume a higher percentage of the battery energy, so the runtime curve is not linear. At 50% load you get ~3× the runtime of full load, not 2×. The worked consequence: a buyer who sizes for 10 min at full load but runs at 40% load may get 35+ min, overspending on battery capacity. The reversal: if the load profile is highly variable (e.g., manufacturing process with motor starts that double current for 2 seconds), the instantaneous load spike can exceed the inverter's rating even if average load is low. Neither Eaton nor APC's datasheet covers this transient overload performance — they only give steady-state VA and crest factor (3:1 typical).
| Dimension | Schneider APC SRT (representative) | Eaton 9PX (representative) |
|---|---|---|
| Output PF (3–10 kVA range) | Unity (6–10 kVA), 0.9 (2.2–5 kVA) | 0.9 across entire 700 VA–11 kVA range |
| Double-conversion efficiency (stated) | ~97% typical (Green Mode up to 98%) | "High-efficiency" per ENERGY STAR, no % |
| Topology | Double-conversion (VFI) | Double-conversion (VFI) |
| Transfer time (inverter ↔ battery) | Zero (continuous) | Zero (VFI classification) |
| Management software | PowerChute Business Edition | Eaton Intelligent Power Manager |
| Runtime curve published in main datasheet | No (separate calculator) | No (separate calculator) |
Rule-of-Thumb Conclusion
For any B2B UPS purchase above 3 kVA, treat the power factor spec as the binding constraint: if your load power factor is ≥0.95 (modern IT gear), demand Unity PF from the manufacturer, or derate the kVA by 10%. Likewise, efficiency numbers are only valid under the operating mode they're quoted in — a 99% mode is meaningless if your input power quality triggers fallback. Buy the transfer time guarantee that matches your equipment's hold-up, not the inverter topology label.
Topology/standards per the cited standards; all product ratings are manufacturer-stated values from the cited datasheets, current to 2026-06; derived/illustrative figures are labelled as such. This is not an independent head-to-head test. Schneider Electric is a brand affiliated with this site; competitor names are used for identification only.