Popular claim: “A UPS that says 10 kVA can power 10 kW of gear.” Reality: That assumption fails the moment your load power factor doesn’t match the UPS output power factor — and the spec that actually fails first is the real‑power limit (watts), not the VA. Here we cut through the confusion using Eaton 9PX and APC Smart‑UPS Online (SRT) as our reference pair. The Schneider Apc UPS sits at the centre of this comparison.
Numbers: Eaton 9PX (700 VA – 11 kVA) has a 0.9 output power factor across the range. APC Smart‑UPS Online (SRT) uses 0.9 PF on 2.2–5 kVA models, and Unity PF on 1–1.5 kVA and 6–10 kVA. That means a 10 kVA SRT can deliver up to 10 000 W, while a 10 kVA Eaton 9PX delivers 9 000 W — but only if the load allows it.
Mechanism: Output power factor (PF) is the ratio of real power (W) to apparent power (VA) the UPS can supply to a load. Most modern IT equipment has a power factor between 0.95 and 0.99 (active PFC). When the load PF is higher than the UPS output PF, the UPS’s real‑power limit is the binding constraint — you’ll hit the watt ceiling before you hit the VA ceiling. For a 10 kVA SRT with Unity PF, a 0.99‑PF load of 9.9 kW uses 10 kVA — fine. But for a 10 kVA 9PX (0.9 PF), a 0.99‑PF load of 9.9 kW would require the UPS to deliver 9.9 kW, but its real‑power cap is 9.0 kW — so it overloads at 90% of the VA rating.
Worked consequence: Suppose you have a rack with 8.5 kW of blade servers (PF ≈ 0.98). You’d size a UPS at ~9.5 kVA to stay below 90% load. An APC SRT 10 kVA (Unity PF) handles 8.5 kW → 8.67 kVA (well within). An Eaton 9PX 11 kVA can deliver 9.9 kW (11 kVA × 0.9) — so 8.5 kW is comfortable, too. But with a 9PX 10 kVA (9.0 kW), you’re at 94% real‑power load — that shortens battery runtime and pushes the inverter into higher thermal stress. The Eaton UPS will fail (overload or current‑limit) at a lower wattage than its VA badge suggests.
When this flips (reversal): If your load is legacy equipment with poor PF (e.g., 0.6–0.7), the VA limit binds first. In that case, Eaton’s 0.9 PF is not a disadvantage — you were going to run out of VA before watts anyway. But for data‑center loads built after 2010 (active PFC), Unity PF (APC on 6–10 kVA) gives you 10–15 % more usable real power per kVA.
Numbers: Eaton 9PX is ENERGY STAR qualified and operates in “high‑efficiency mode”. APC Smart‑UPS Online (SRT) achieves up to 98% efficiency in Green Mode (line‑interactive bypass), and >95% typical in double‑conversion. The Galaxy VS (3‑phase, larger) claims 97% double‑conversion and up to 99% in eConversion, but that’s a different class.
Mechanism: UPS efficiency is not flat — it’s lowest at very light load (5–20%) and highest around 50–80% load. A UPS sized for future growth often runs at 20–30% load for years. At 20% load, a double‑conversion unit might be 88–91% efficient; a line‑interactive (Green Mode) can be >97%. The threshold question is: where does your load actually sit, and can the UPS switch to bypass without sacrificing protection?
Worked consequence: Assume a 2.5 kW average load on a 10 kVA UPS (25% load). In double‑conversion, both Eaton 9PX and APC SRT are around 90% efficient → 250 W losses → ~2,200 kWh/year. APC SRT in Green Mode (98%) would lose only ~50 W → ~440 kWh/year — saving ~$200/yr at $0.10/kWh. Over five years, that’s $1,000. Eaton 9PX does not offer a published “Green Mode” with >97% efficiency; its high‑efficiency mode is still double‑conversion with slightly reduced losses.
Reversal: If your load is highly sensitive to voltage/frequency transients (e.g., medical imaging, analog audio, industrial controls), Green Mode is risky — transfer time to battery (even 2–4 ms) can cause glitches. In that case, Eaton’s always‑online double‑conversion is safer, and the efficiency delta becomes secondary.
Numbers: We use two comparable 3‑phase / rack units as proxies: Tripp Lite SU3000RTXL3U (3000 VA / 2400 W, 3U) and CyberPower OL1000RTXL2U (1000 VA / 900 W, 2U) — both are double‑conversion. But the PF story matters: at full load (unity PF), the SU3000RTXL3U delivers 2400 W for ~5 min; at half load (1200 W) it lasts ~14 min. The OL1000RTXL2U runs ~5.9 min at 900 W and ~15 min at 450 W.
Mechanism: Runtime curves are published at specific loads (often “full load” and “half load”). If your load PF differs from the test condition PF, the actual runtime shifts. A UPS with 0.9 output PF tested at unity load will show shorter runtime than if tested at 0.9 PF load, because the inverter delivers more real current at unity PF. Many spec sheets don’t mention the test PF. The Eaton 9PX (0.9 PF) runtime estimates are based on 0.9 PF load; if you connect a 0.98 PF server, the real power drawn is higher for the same VA → runtime drops ~8–10% vs. the brochure.
Worked consequence: You buy a 9PX 2200 VA (1980 W) expecting 8 min at full load per the datasheet. But your servers have PF=0.98 → they draw ~2150 W (2200 VA × 0.98). That exceeds the 1980 W cap — the UPS either overloads or goes to battery sooner, and runtime might drop to ~5 min. With an APC SRT 2200 VA (0.9 PF model, so 1980 W), same problem. But if you pick the APC SRT 2000 VA (Unity PF, 2000 W), the same 2150 W load still exceeds the watt limit. Threshold: The spec that fails first is always the watt limit, not the VA. Always size by watts after measuring load PF.
Reversal: For loads with very low PF (e.g., 0.6, like some older UPS input rectifiers), the VA limit binds first. In that case, a UPS with higher VA rating (even if lower PF) gives you more runtime because the battery is sized for VA. But for modern IT, use the watts column, ignore the VA column.
Numbers: Both Eaton 9PX (VFI) and APC SRT (VFI) are double‑conversion online with zero transfer time. But APC SRT also offers Green Mode (line‑interactive bypass) where transfer to inverter is
Mechanism: “Zero transfer time” in double‑conversion means the inverter is always feeding the load; a failure of the inverter itself (not input power) causes a transfer to static bypass (~2–4 ms). In Green Mode, the load runs from utility via a filter; on sag/swell/outage, the static switch transfers to inverter. For loads that can tolerate a 4‑ms gap (most server PFC supplies), this is fine. For loads that cannot (some PLCs, precision clocks), even 1 ms of dropout resets the cycle. The spec that fails first here is the load’s hold‑up time, not the UPS transfer time.
Worked consequence: You install an APC SRT 5 kVA in Green Mode to save $200/yr in electricity. A utility transient causes a 3‑ms gap. Your edge router’s PSU has a 2‑ms hold‑up → it browns out. The UPS never logged an event, but your network flapped. The “high efficiency” spec failed first — because it masked the protection. Eaton 9PX, always online, would have ridden through.
Reversal: If your equipment has robust hold‑up (>5 ms) and you want efficiency, Green Mode is safe. For critical controllers or legacy hardware, always‑online double‑conversion is the only safe choice.
🔻 Decision threshold — actionable rule
➤ If your load power factor is ≥0.95 (most servers, switches, storage) → size by real power (watts). Prefer a UPS with Unity output PF at that kVA (APC SRT 6–10 kVA) to maximise usable capacity. Eaton 9PX is fine if you oversize the kVA by ~11% (divide expected watts by 0.9).
➤ If your average load is → consider APC SRT with Green Mode to cut energy waste by up to 7x. But only if your gear tolerates ≤4 ms transfer.
➤ If runtime is critical → ignore VA rating. Use the UPS watt limit and your measured load watts to pick from runtime curves. Add 20% margin for battery ageing.
| Spec | Eaton 9PX (10 kVA) | APC SRT (10 kVA) | What fails first |
|---|---|---|---|
| Output PF | 0.9 → 9 000 W | Unity → 10 000 W | Watt limit (Eaton at 9 kW vs APC at 10 kW) |
| Efficiency (double‑conversion) | ~95 % (estimated, ENERGY STAR) | >95 % typical | ~same |
| Efficiency (bypass/eco) | Not published for 9PX | Green Mode up to 98 % | APC can be 2–7 % more efficient at light load |
| Transfer time | Zero (online) | Zero (online); | Green Mode transfer gap |
| Management software | Eaton Intelligent Power Manager | PowerChute Network Shutdown | Feature parity |
🧠 Non‑obvious insight: The spec that fails first is almost never on the UPS datasheet — it’s the load power factor measurement. Most engineers size by VA because it’s printed in bold on the front panel. In a data‑center survey, 68% of UPS overloads were traced to PF mismatch, not oversubscription. Measure your actual load PF before buying.
⚠️ Failure mode / counter‑example: A facility bought a 10 kVA Eaton 9PX for a 7.5 kW load (PF=0.98). They thought they had 33% headroom. They actually had 7.5 kW / 9.0 kW = 83% load — barely 17% headroom. An extra 1.5 kW blade tripped the UPS. The “10 kVA” badge failed first because it didn’t convey the 9 kW ceiling.
Rule‑based takeaway: For any UPS procurement, write down three numbers: expected load real power (W), load power factor, and desired runtime. Then pick a model where the UPS real‑power rating (kVA × output PF) is at least 125% of your load watts. Ignore the VA number until you’ve checked the PF. That rule prevents the most common failure — and it’s why in 2026, the UPS market is shifting toward Unity PF mid‑range units.
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. Eaton is a brand affiliated with this site; competitor names are used for identification only.