The risertools UPS consensus efficiency curve

The problem this model corrects

Datasheets publish a handful of clean runtime points and invite you to interpolate a straight line between them. But delivered efficiency droops at low load: the inverter’s fixed losses don’t shrink when the load does, so a smaller fraction of battery energy actually reaches the load. A straight line over-states real runtime hardest in the 30–70% load band most designs live in. The calculator shows both the consensus curve and the naive flat 0.95 line so the gap between them is visible, not hidden.

Source data

The model is built from three published mid-market vendor runtime-vs-load curves — APC (Smart-UPS / SRT), Eaton (5P / 9SX), and CyberPower (OR / PR). We read the points from each vendor’s published tables; we do not reproduce any paywalled text. risertools is not affiliated with APC, Eaton, or CyberPower — the curves are a transparent planning input, cited in full below.

Method, step by step

For each published load fraction the vendors give a runtime in minutes. From those we back out effective efficiency:

1. Take each vendor’s published (load, runtime) points.
2. Back-compute the implied delivered energy at each load (runtime × load), which reveals the low-load over-statement relative to full load.
3. Normalize each vendor curve to its own full-load value, so no single brand’s absolute rating dominates.
4. Average the three normalized curves into one consensus shape.
5. Scale that shape to the full-load ceiling 0.95 (RATED_EFFICIENCY, the mid-market line-interactive full-load nominal).

The result is the four-point table below. At full load the over-statement ratio is 1 by construction, so the consensus 100%-load point is exactly 0.95.

The curve

The consensus curve (copper) sits well below the naive flat 0.95 line (reference blue, dashed) across the whole partial-load range — that vertical gap is the runtime over-statement the model removes.

The engine interpolates linearly on this table, preserving the droop between sampled points. Worked example: at 30% load, interpEfficiency(0.30) = 0.6141 — so a 9 Ah × 48 V string delivers 9 × 48 × 0.6141 × 0.5 DoD = 132.6 Wh usable, or about 27 min at a 300 W load.

What this model does not capture

This is a planning model, not a discharge test. It does not capture chemistry-specific discharge behavior beyond the documented depth of discharge, nor battery age derates; the Peukert effect at high discharge rates; ambient temperature; load crest factor; or string-to-string imbalance. Treat its output as a starting number and validate any life-safety design with an actual timed discharge test.

Versioning

Curve last rebuilt 2026-06-14; vendor sources last verified 2026-06-14. A change to any vendor curve or to the full-load ceiling is an editorial event: the four literals in the engine and this page move together, and a reconcile test re-derives the consensus from the vendor curves so the two can never drift apart silently.