You buy a UPS for the watts it can actually supply to your load, not the VA number on the sticker. Yet most spec sheets lead with VA, and the watt rating depends on the output power factor (PF). The difference between a 0.7 PF unit and a 0.9 PF unit at the same VA can mean 20–30% less usable wattage — which either strands capacity or forces you to overspend. This tear-down compares three tangible dimensions where real-watt sizing changes the decision, using Schneider Electric APC Smart-UPS Online (SRT) and Eaton 9PX as the primary pair. Every number is from manufacturer datasheets; derived figures are labelled.
1. Output Power Factor: The Real-Watt Multiplier
Numbers. Schneider APC Smart-UPS Online (SRT) in the 2.2–5 kVA range offers an output power factor of 0.9; the 1–1.5 kVA and 6–10 kVA models deliver unity PF (1.0). Eaton 9PX across its 700 VA–11 kVA range uses a 0.9 output PF. So at 3000 VA, Eaton 9PX gives 2700 W (3000 × 0.9). A Schneider SRT at 3000 VA in the 2.2–5 kVA band also gives 2700 W. If you compare an older or lower-spec UPS that uses 0.7 PF, the same 3000 VA yields only 2100 W — a 22% haircut in real capacity.
Mechanism. Output PF defines the ratio of real power (watts) to apparent power (VA) the inverter can deliver to the load. A higher PF means the inverter can handle more resistive/rectifier load for the same apparent rating. This is not an efficiency figure; it is a capability limit imposed by the inverter bridge and output filter design. A unity PF unit can deliver its full VA as watts, which matters when you power modern switch-mode PSUs that have a PF of 0.95–0.99.
Worked consequence. Assume a rack of three switches, each rated 800 W (2400 W total). With a 0.9 PF UPS at 3000 VA you get 2700 W — margin of 300 W, which is tight but acceptable. With a 0.7 PF UPS at 3000 VA you get only 2100 W — under-sized by 300 W, forcing you to buy a 4000 VA unit (2800 W) at roughly 33% more cost. For a 2400 W load, the break point is clear: you need at least 2667 VA at 0.9 PF (Schneider SRT or Eaton 9PX at 3 kVA) versus 3429 VA at 0.7 PF.
When it flips. If your load is purely resistive (heating, incandescent lighting) or uses legacy linear PSUs with PF near 0.6–0.7, the higher PF rating gives no benefit — the inverter still delivers full VA, but the load won't pull more watts than its PF allows. In data centers with modern PFC power supplies, the high PF is always an advantage.
2. Efficiency Under Real Load: Where the Losses Actually Are
Numbers. Schneider APC Smart-UPS Online (SRT) offers double-conversion efficiency at ~94–96% (typical) and a Green Mode that can reach up to 98%. Eaton 9PX is described as high-efficiency operation, ENERGY STAR qualified. For the 3-phase class, Schneider Galaxy VS delivers double-conversion efficiency up to 97% at every load level, and an eConversion mode up to 99% with Class 1 performance and no-break transfer.
Mechanism. Efficiency in double-conversion mode is the ratio of AC-out to AC-in, and the ~4–6% loss is mostly heat from the rectifier and inverter. But the more important number is how efficiency behaves at partial load. Many UPS units are optimised for 60–80% load and drop below 90% at 10–20% load. Both Schneider SRT and Eaton 9PX claim flat efficiency curves; however, the Green/eConversion modes bypass the inverter when the grid is stable, cutting losses by roughly 2–4 percentage points. This matters because the average load in a lightly-loaded rack may be only 30–40% of the UPS rating.
Worked consequence. For a 10 kVA Schneider SRT in double-conversion, losing 4% = 400 W of heat that your cooling must remove. At $0.12/kWh, that's ~$421/year per UPS just in electrical loss. Switching to eConversion (99%) reduces loss to ~100 W, saving ~$315/year. Eaton 9PX in double-conversion at similar rating would produce ~400–500 W of heat depending on load. Over a 5-year lifecycle, the cumulative energy-cost difference between a 96% and 98% efficient UPS is roughly $300–$800 per unit, which can exceed the hardware price delta between brands.
When it flips. If your site has very unstable power (frequent brownouts, voltage sags), running in eConversion/Green Mode may risk an overload transfer if the inverter cannot switch fast enough. Schneider UPS's eConversion claims no-break transfer and Class 1 performance; Eaton UPS does not publish equivalent mode specs for 9PX. For mission-critical loads with poor power quality, you must stay in double-conversion despite the efficiency penalty — then efficiency differences narrow to 1–2%, which is less decisive for small UPS.
3. Power Density and Heat Rejection: Space vs Cooling
Numbers. Eaton 9PX delivers up to 5400 W in 3U, and 10 kW in 6U rack/tower. Schneider APC SRT in the 6–10 kVA range delivers up to 10,000 W in 6U. For the 3 kW class, both are typically 2U (Schneider SRT 3000 VA) vs Eaton 9PX 3000 VA in 2U. So watt-per-U is comparable across brands at similar ratings.
Mechanism. Power density (watts per rack unit) matters for floor space, but the corollary is heat rejection. The heat a UPS rejects to the room is roughly: (1 – efficiency) × real load. At 96% efficiency and 2400 W load, heat = 0.04 × 2400 = 96 W. At 98% efficiency, heat = 48 W. The density difference between a 2U unit (5400 W in 3U = 1800 W/U) and a 3U unit (3000 W in 2U = 1500 W/U) is not a heat rejection difference — both reject ~100 W heat for the same load, but the 3U unit has more surface area for cooling. In a dense rack, a 2U unit may run hotter internal components, potentially affecting reliability at high ambient temperatures.
Worked consequence. If you have a 42U rack and want to maximise compute density, a 2U UPS saves 1U per unit. Over 10 racks, that's 10U of extra compute space — worth ~$2,000–$5,000 in colo space per rack per year. But if the 2U unit's fans run at high speed to shed heat, the noise and MTBF may be worse. For typical B2B closets with 2–4 racks, the space saving is minimal; the heat load is dictated by the load, not the UPS volume.
When it flips. In a high-density compute scenario where every U counts, the denser unit wins regardless of brand. But for a lightly loaded IT room, a physically larger unit may cost less and run quieter. Neither Schneider nor Eaton has a blanket advantage here — you must compare specific models in your power band.
Magnitude Proportion: The Decisive Ratios
Here's the condensed arithmetic:
- Watt capacity ratio between 0.9 PF and 0.7 PF at same VA: 1.29× (29% more real watts). That's the multiplier that matters.
- Efficiency improvement from 96% to 98%: ratio of 1.02× in watts delivered per watt in, but the loss halves — from 4% to 2% — which is a 50% reduction in heat loss.
- Cooling cost ratio: each watt of heat saved reduces AC load by ~0.4 W in a typical data center (using PUE ~1.4). So saving 200 W of heat saves ~80 W of cooling power, worth ~$70/year at $0.12/kWh.
The dominant term in the lifecycle cost equation is the real-watt vs VA mismatch — a 29% capacity swing that can force a tier jump. Efficiency and density are secondary but not negligible.
Failure Mode / When This Breaks Down
If you compare the Schneider SRT in Green Mode (98%) vs Eaton 9PX in double-conversion (say 94%), the efficiency gap is 4 percentage points. But if your power quality is poor and you cannot use Green Mode, both units run double-conversion at ~95–96%. At that point the PF difference is irrelevant because both are 0.9 PF. The only remaining differentiator is management software (Schneider PowerChute vs Eaton IPM/software, not covered in this comparison). Rule of thumb: if your site has clean utility power and you can use high-efficiency mode, a Schneider SRT with Green Mode saves meaningful kWh. If your site is dirty or you are locked into double-conversion, the real-watt advantage from PF is the only solid edge — and both brands offer it equally.
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.