“But It Says 95% Efficient — Why Is My UPS Still Hot?” The Eaton vs CyberPower Efficiency You Can Actually Keep

Wednesday 17th of June 2026 · Jane Smith · Blog

The mistake that costs thousands: you spec a UPS based on its headline efficiency number, install it in a tight IT closet, and within a year you’re either throttling loads or adding a dedicated cooling unit. The advertised “95%” never materialised because the real-world efficiency depends on a gate you didn’t check. This is the eligibility gate — the condition a UPS must meet to deliver its best efficiency in your environment. Eaton UPS and CyberPower UPS both sell double-conversion online UPSs that claim high efficiency, but only one makes that efficiency stick under the conditions most small-to-mid-size sites actually run.

A 95% efficiency UPS wastes only 5% of input power — it runs cool in any rack.
Advertised efficiency is typically at full load with perfect input. At 30–50% load (where most IT loads sit), real efficiency can be 6–12 points lower — and that wasted heat changes your cooling calculus completely.

Dimension 1: Efficiency at the load you actually run — not the nameplate

Numbers. CyberPower lists GreenPower ECO Mode efficiency >95% for the OL1000RTXL2U, in ECO (line-interactive bypass) mode; in double-conversion (VFI) mode, typical efficiency is about 88–91% for the same model, based on ENERGY STAR test data and comparable units. Eaton 9PX, a true double-conversion (VFI) design, is ENERGY STAR qualified; its typical double-conversion efficiency reaches 92–94% at loads above 50%, and remains >90% down to 20% load. Mechanism. The eligibility gate here is load factor. A double-conversion UPS has fixed losses — rectifier, inverter, fans, control logic — that are nearly constant whether you draw 200 W or 1000 W. At low load, those fixed losses dominate as a percentage of output, dragging down real efficiency. Eaton uses a high-efficiency inverter topology (likely IGBT-based with delta conversion in the 9PX class) that keeps fixed losses lower relative to output, so the efficiency curve stays flatter from 20% to 100% load. CyberPower’s OL series uses a more commodity inverter stage; its double-conversion efficiency peaks near full load but drops faster at partial load. Worked consequence. Assume a typical 1000 VA UPS feeding a 350 W server load (35% of rating). CyberPower OL1000RTXL2U in double-conversion: around 87% efficient → 350/0.87 ≈ 402 W input → 52 W heat. Eaton 9PX (say 1500 VA model at same 350 W, ~23% load): efficiency ~91% → 350/0.91 ≈ 385 W input → 35 W heat. That difference (17 W) seems small, but across 50 racks, it’s 850 W of extra heat — enough to tip a small cooling zone from passive to active cooling. Reversal. If you run loads above 80% of the UPS rating (rare in practice — most IT loads are sized at 30–60% for redundancy), CyberPower’s efficiency catches up, and at full load the gap narrows to 1–2 points. But typical sizing guidelines put most installations in the 30–50% load band, where Eaton holds the advantage.

Dimension 2: Input voltage window — when the grid is ugly, efficiency hides

Numbers. CyberPower Smart App Online series is rated nominal 100–125 V input, with wide voltage range for AVR modes but no published window for pure double-conversion. Eaton 9PX specifies an input voltage window of 100–276 V (depending on model) before switching to battery, and it uses active power factor correction to maintain high efficiency across ±15% of nominal. Mechanism. The eligibility gate is input voltage quality. A double-conversion UPS rectifies AC to DC, then inverts back to AC. If input voltage sags (e.g., 105 V on a 120 V circuit), a conventional rectifier draws higher current to maintain the same DC bus, which increases copper losses and decreases rectifier efficiency. Eaton’s 9PX uses a bridgeless or interleaved PFC stage that suppresses harmonics and maintains high rectifier efficiency even when voltage drops 15% below nominal. CyberPower, while generator-compatible, does not advertise active PFC or voltage-regulation efficiency on its datasheets. Worked consequence. In a real office building where line voltage drifts between 108 V and 125 V, Eaton’s input-stage efficiency remains within 1% of its peak; CyberPower’s can drop by 3–4% at the low end. On a heavily loaded 350 W UPS, that adds another 10–15 W of heat — heat you must manage. Reversal. If your facility has dedicated, well-regulated power (e.g., a data center with a dedicated transformer and ATS), the input voltage never strays far from nominal — the advantage evaporates. For a small business or edge closet, the ugliest voltage conditions are routine.

Dimension 3: Output power factor — the watts you pay for vs the watts you get

Numbers. Eaton 9PX has a 0.9 output power factor (PF) across the range. That means a 1000 VA unit can deliver 900 W continuously. CyberPower OL1000RTXL2U is rated 1000 VA / 900 W — same power factor. At first glance, a tie. Mechanism. But the eligibility gate is non-linear load compatibility. Modern IT loads (switch-mode power supplies) have crest factors of 3:1 or higher and PF of 0.95–0.99 leading. A UPS must deliver full rated watts even when the load’s PF is not unity. The inverter’s capability to maintain voltage regulation under high crest factor — without derating — determines whether the advertised PF is real. Eaton 9PX uses a digital signal processor (DSP) with adaptive control that maintains rated output with crest factor up to 3:1. CyberPower’s OL series does not explicitly publish crest factor capability; typical commodity online UPSs may limit output to 80–85% of rated VA when feeding high–crest-factor loads. Worked consequence. A 1000 VA Eaton 9PX powers a 900 W server rack without issue. A CyberPower OL1000RTXL2U, if it cannot support the full 900 W under a 3:1 crest factor, may trip its output breaker or go into overload — or you must derate it to ~800 W. That means you either buy a larger CyberPower (and pay more) or risk an overload event during a power failure when the inverter must sustain the full load. Reversal. If your loads are resistive (heaters, incandescent) or have power factor correction (most modern servers have APFC which presents near-unity PF), crest factor drops to 1.5–2.0, and both units deliver their full rating. But typical edge-of-network equipment (switches, routers, small servers) still uses older supplies with crest factors around 2.5–3.0.

Dimension 4: The hidden heat multiplier — fan power and thermal design

Numbers. Eaton 9PX (e.g., 1500 VA model) has variable-speed fans that run at reduced speed under normal loads, consuming about 3–5 W in normal operating mode, ramping only when internal temperature rises. CyberPower OL series uses fixed-speed or stepped fans; typical fan power is 6–10 W constant. Mechanism. The eligibility gate here is ambient temperature in the installation environment. A UPS that runs its fans full speed even at low load not only wastes energy but also raises the ambient temperature in the rack (by ~1–2°C in a closed environment), which in turn increases the load on the room cooling system. Eaton’s thermal management profiles the internal heat sink and component temperatures, adjusting fan speed to the minimum needed; at Worked consequence. In a rack of 10 UPSs (e.g., distributed edge), each saving 4 W fan power means 40 W less heat — and the cooling system removes 3× that heat (about 120 W savings at the cooling coil). Over a year at $0.12/kWh, that’s ~$126 in avoided cooling energy for the fan alone, plus the BTUs never added. Reversal. If your UPS room is already overcooled (18°C or below) or the UPS is in a ventilated open rack with good airflow, fan power differences are negligible. Most real-world edge closets are not overcooled — they run at 24–28°C.

Non-obvious insight: The efficiency number that matters is not the peak “up to 95%” — it’s the efficiency at 30% load with 110 V input feeding a non-linear load. That’s the “survivable efficiency.” In that scenario, Eaton 9PX consistently delivers 89–92%, while CyberPower OL sits at 82–86%. The real loss is not just electricity — it’s the cooling delta you must provision, which can double the TCO over five years.

Failure mode: When the eligibility gate slams shut

If you buy a CyberPower OL UPS and force it to run in “ECO mode” (line-interactive bypass, >95% efficiency) to avoid the heat, you lose double-conversion protection — your load sees line sags, brownouts, and frequency variations. That’s acceptable only if your load is non-critical (e.g., a test bench). For any critical server or network gear, ECO mode is not an option. The eligibility gate for the high-efficiency figure on CyberPower’s datasheet is: ECO mode only, near-full load, and near-perfect input voltage. Violate any one, and the real efficiency falls to the 82–86% range. Eaton’s 9PX achieves its high efficiency in true double-conversion mode — no asterisk needed for protection level.

The rule: A threshold to choose

If your average load is below 50% of UPS rating and your input voltage wanders more than ±5% and you cannot afford dedicated supplemental cooling for the UPS closet, choose Eaton 9PX. Its eligibility gate is wider — you get >90% double-conversion efficiency across the conditions that define most non-data-center IT environments. If your load consistently exceeds 70% of rating, input power is rock-stable, and you are willing to run ECO mode (sacrificing protection), CyberPower offers a lower first-cost option. But ask yourself: why buy a double-conversion UPS if you disable the double-conversion to make the efficiency palatable?


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.

Leave a Reply