UPS Runtime Calculator

Your system load

Total system draw (watts)

Runtime target UPS type

Your UPS (if known)

UPS preset UPS VA rating

UPS watt rating

Runtime estimate

— Est. minutes

— % of capacity

— Headroom (W)

Load vs UPS capacity

0%50%80%100%

Runtime at different load levels (your UPS)

System load	% of UPS watts	Est. runtime	Assessment

Runtime is a rough estimate: consumer sealed-lead-acid UPS hold about 0.14 Wh per VA (a 1500 VA unit ≈ ~210 Wh), adjusted for inverter efficiency and a load-dependent capacity derate. The runtime model is calibrated to typical consumer line-interactive UPS curves; larger-battery online / rackmount units run longer, so it errs low for those. Actual runtime varies with battery age (new batteries outlast aged by 15–30%), temperature, and model, so check your specific UPS model's published runtime chart for exact figures before sizing critical gear.

What size UPS do you need?

Worked examples

These are this calculator’s own outputs, computed with the same formula the tool runs in your browser — so you can see a real answer without touching a single input.

Tower server + NAS + switch drawing 150W, on a CyberPower CP1500PFCLCD (1500VA / 1000W, line-interactive). This is the calculator’s own default state

Inputs: Total system draw = 150 W; UPS VA rating = 1500 VA; UPS watt rating = 1000 W; UPS type = Line-interactive (PF 0.9); Runtime target = 10 minutes
Result: 32 min | 15% of capacity | +850W headroom. Status: green (“Good load level - below 50% of rated capacity”). Meets the 10-minute target. Recommendation cards at 150W: CP1000PFCLCD 20 min, CP1500PFCLCD 32 min, PR2200LCDRT2U 48 min - all three earn the “Fits” badge.

Small homelab - 2x mini PC + NAS + switch drawing 80W, on a CyberPower CP1000PFCLCD (1000VA / 600W, line-interactive)

Inputs: Total system draw = 80 W; UPS VA rating = 1000 VA; UPS watt rating = 600 W; UPS type = Line-interactive (PF 0.9); Runtime target = 10 minutes
Result: 41 min | 13% of capacity | +520W headroom. Status: green. Comfortably meets the 10-minute target. Recommendation cards at 80W: CP1000PFCLCD 41 min, CP1500PFCLCD 63 min, PR2200LCDRT2U 91 min.

Tower + GPU + NAS + network gear drawing 350W, on a CyberPower CP1500PFCLCD (1500VA / 1000W, line-interactive)

Inputs: Total system draw = 350 W; UPS VA rating = 1500 VA; UPS watt rating = 1000 W; UPS type = Line-interactive (PF 0.9); Runtime target = 10 minutes
Result: 12 min | 35% of capacity | +650W headroom. Status: green (still under 50%). Meets the 10-minute target with ~2 min margin. Recommendation cards at 350W: CP1000PFCLCD 7 min (no badge - misses the 10-min target), CP1500PFCLCD 12 min (Fits), PR2200LCDRT2U 19 min (Fits).

How to use this calculator

Enter your total system load in watts — use your actual measured draw or the output from the Power & Cost Calculator if you’ve already built your system profile there. Then either select a UPS model from the presets or enter your UPS VA and watt ratings manually.

The runtime estimate assumes a standard sealed lead-acid battery in good condition (under 2 years old). Battery age is the largest variable in real-world UPS performance.

Understanding the results

Est. minutes is calculated using the standard lead-acid energy model: battery capacity (Wh) × inverter efficiency ÷ load (W), adjusted for the Peukert effect at higher load fractions. The usable fraction of nameplate capacity is always well below 100% - the inverter’s low-voltage cutoff strands part of the pack no matter how light the load - and it shrinks further as load rises. The model runs from roughly 45% of nameplate watt-hours at very light loads down to about 24% at the UPS’s full watt rating.

% of capacity is your load divided by the UPS watt rating. This is the number to watch. Stay under 80%; aim for under 50%.

Headroom is watts remaining before you hit the UPS’s rated limit. This matters for inrush current — hard drives and GPUs can draw 2–3× their steady-state wattage at startup. A UPS at 90% capacity during steady-state may trip at startup of a large GPU.

Sizing for a Proxmox homelab

The rule I use: load the UPS to 50% of rated watts, target 10 minutes of runtime. That gives a comfortable shutdown window when combined with NUT (Network UPS Tools) configured for automatic VM shutdown.

Homelab config	Idle draw	Recommended UPS
N100 mini PC + switch	20–30W	850VA / 510W (CP850PFCLCD)
2× mini PC + NAS + switch	60–90W	1000VA / 600W (CP1000PFCLCD)
Tower server + NAS + switch	150–220W	1500VA / 1000W (CP1500PFCLCD)
Tower + GPU + NAS + network	280–400W	2150VA / 1980W (PR2200LCDRT2U)

The CyberPower PFC Sinewave series (PFCLCD suffix) and the Smart App Sinewave series (PR prefix) are both listed by CyberPower with “Waveform: Sine Wave” — required for active PFC power supplies. Check the waveform line on the spec sheet before buying: CyberPower’s own catalog mixes true-sinewave and “Simulated Sine Wave” units within the same VA class (the 2100VA OR2200LCDRTXL2U is simulated, the 2150VA PR2200LCDRT2U is not). Do not buy a simulated-sine UPS for homelab hardware.

Battery replacement math

The battery is a consumable. CyberPower and APC replacement batteries run $25–50 for the 1500VA class and typically need replacement every 3–5 years. Factor that into your TCO: a $150 UPS plus one battery replacement over 5 years is a $175–200 5-year cost, not $150.

You can verify battery condition any time with upsc (NUT) or your UPS’s self-test button. A UPS that fails its self-test or reports low battery capacity is overdue for a replacement cell.

What this calculator does not cover

This is a runtime estimator for standard sealed lead-acid (VRLA) UPS units. Lithium-ion UPS units (CyberPower’s LX series, APC’s BX-Li series) have different discharge curves — flatter, higher capacity-to-weight — and will outperform the estimates here at the same VA rating. Extended-runtime models with external battery packs (APC SMT1500RM2U with APCRBC) are also outside the scope of this calculator’s model.

For the full sizing methodology including VA/W math, line-interactive vs online double-conversion tradeoffs, NUT configuration, and battery replacement scheduling, see the UPS and Power Sizing guide.

Corrections

2026-08-14 — two recommended UPS models were not real, and one watt rating was wrong. This page previously recommended a “CyberPower CP2200PFCLCD (2200VA / 1320W)” in the preset list, the recommendation cards, the sizing table and all three worked examples, and a “CP650E” in the sizing table. Neither model exists: CyberPower’s own site search returns 0 results for each, and the PFC Sinewave line ends at the CP1500PFCLCD. They are now the PR2200LCDRT2U (2150VA / 1980W) and the CP850PFCLCD (850VA / 510W), both listed by CyberPower with “Waveform: Sine Wave.” The CP1500PFCLCD was also listed here at 900W; CyberPower publishes 1500VA / 1000W. Every figure that depended on that rating - the calculator default, the percent-of-capacity and headroom numbers in the worked examples, the sizing table - has been recomputed. If you sized a UPS from this page before 2026-08-14 using the CP2200PFCLCD row, that purchase decision was based on a product you cannot buy; re-run the calculator.

Assumptions and sources

The runtime model - battery watt-hours times inverter efficiency divided by load, with usable capacity shrinking at higher load fractions - is the standard lead-acid estimation method described in Battery University’s battery-runtime guide.
The UPS preset ratings (for example the CyberPower CP1500PFCLCD’s 1500 VA / 1000 W) are the manufacturer’s published specifications; see CyberPower’s CP1500PFCLCD product page, which also publishes that unit’s own runtime figures (10 minutes at half load, 2.5 minutes at full load). Those are the anchors this calculator’s discharge curve is checked against, and it does not pass both: it reproduces the full-load figure (2.7 minutes modelled against 2.5 published) and reads about 21% low at half load (7.9 against 10). The curve the coefficients were originally fitted to was retracted on 2026-08-25 as not vendor-published, and they were left as they are rather than refitted to these two points, because a two-parameter fit to two points can no longer be checked by anything and would move the model in the optimistic direction on a tool whose failure mode is buying too little battery.
The battery-capacity figure behind each preset (0.14 Wh per VA) and the accuracy bands quoted in the FAQ are TechFuelHQ modeling assumptions for a healthy battery at room temperature, stated so you can derate for age and cold - the two effects the FAQ quantifies. The 0.14 Wh/VA ratio is read off consumer line-interactive towers (APC BR1500MS2 216 Wh / 1500 VA, APC BE650 84 Wh / 650 VA, CyberPower CP1500PFCLCD 216 Wh / 1500 VA), which is why the model reads low for rackmount and online units with larger banks.
The overload test applies a load power factor of 0.95 - the conservative end of the 0.95-0.99 that the FAQ above describes for PFC computer supplies - to convert your watts into the VA the UPS has to supply, and rejects a load that exceeds either the unit’s watt rating or its VA rating. It previously divided by the UPS’s own output power factor instead, which is a property of the UPS rather than of your load, and that rejected loads at and just below some units’ rated watts.

Frequently asked questions

How accurate is the UPS runtime estimate?

It depends on the class of unit, and the model errs low rather than high in every case we can check it against. For the consumer line-interactive towers it is fitted to — the 1500VA-class CyberPower and APC units in the preset list — it lands within about 20% of the vendor’s own published figures: against CyberPower’s CP1500PFCLCD it returns 2.7 minutes at full load against a published 2.5, and 7.9 minutes at half load against a published 10, so roughly +8% and -21%. For units with a larger battery bank than that ratio assumes it reads materially low. Against CyberPower’s PR2200LCDRT2U rackmount it returns 5.7 minutes at half load against a published 9, about -36%. Against APC’s tower SMT1500, whose own published curve runs 1:24 at 200W, 0:23 at 500W, 0:12 at 700W and 0:07 at 1000W, it reads 59-72% low across the whole curve. If your unit publishes its own runtime figures, they beat this estimate every time. Runtime also degrades as batteries age — most sealed lead-acid UPS batteries lose 15–30% capacity by year 3 and need replacement around years 3–5 — and cold below 50°F can cost another 10–20%. Treat these as planning floors, not guarantees, and use your own model’s published runtime chart when you have one.

What percentage of UPS capacity should my load be?

30–50% of rated watts is the sweet spot — you get good runtime, reasonable battery longevity, and headroom for startup inrush current. Above 80% of rated watts, runtime drops sharply and the battery cycles harder. Below 25%, some UPS units have efficiency drops in the inverter stage.

How long do I actually need a UPS to run?

For a homelab running Proxmox with NUT integration: 5–7 minutes is enough for the UPS to signal low battery, trigger a graceful VM shutdown sequence, and allow the system to power off cleanly. 10–15 minutes gives comfortable margin if you want time to investigate before shutting down. Only target 30+ minutes if you need the homelab to keep running through brief outages without any shutdown.

What’s VA vs watts on a UPS?

VA (volt-amperes) is apparent power; watts is real power. The difference is the power factor of your load. Modern PC power supplies have a power factor corrected (PFC) input stage and draw close to unity PF, so watts ≈ VA × 0.95–0.99. The UPS watt rating is the practical limit; VA is the electrical sizing spec. Always size to the watt rating, not VA, when checking if your load fits.

Do I need a pure sine wave UPS for a homelab server?

Yes for any modern UPS. Active PFC power supplies (which all current ATX PSUs use) require a pure sine wave input — they will misbehave, make noise, or shut off on a stepped/simulated sine wave UPS. Every CyberPower unit in this calculator’s preset list is published by the manufacturer as ‘Waveform: Sine Wave.’ The one APC Back-UPS entry (BE650G1) is a budget standby unit and TechFuelHQ has not been able to confirm its waveform against APC’s own spec sheet, so treat it as unconfirmed rather than assumed-safe. Avoid any UPS that advertises ‘stepped approximated’ or ‘simulated’ sine wave.

Evidence ledger

Last updated: August 30, 2026
Methodology: See our methodology for research and review standards.
Update log: 2026-08-30 — Settled the SMT1500 row the 2026-08-24 entry below left open, and corrected the overload guard. (1) The APC SMT1500 preset carried 1500VA / 1050W, a figure that appears on no APC sheet; apc.com and se.com return HTTP 403 to a scripted fetch, which is why the earlier pass could not close it. Reached it through Schneider Electric’s own document server instead - download.schneider-electric.com, p_Doc_Ref=SPD_VYOG-8WLK6U_EN, VYOG-8WLK6U_R4_EN.pdf, HTTP 200, 1,301,031 bytes - whose ‘APC Smart-UPS // Standard Tower Technical Specifications’ table reads verbatim ‘Power Capacity 500W / 750VA 700W / 1000VA 1000W / 1500VA 1980W / 2200VA 2700W / 3000VA’ under the headers ‘SMT750 SMT1000 SMT1500 SMT2200 SMT3000’. The tower SMT1500 is 1500VA / 1000W; the preset value and label are corrected, and the 50W overstatement is out of the watt cutoff and headroom math. Note that se.com lists SMT1500 as discontinued on 2024-06-30, replaced by the SMT1500C. (2) The same sheet publishes this unit’s runtime curve - 1:24 at 200W, 0:23 at 500W, 0:12 at 700W, 0:07 at 1000W - and the model reads 59-72% low against all four points, because the SMT1500’s battery bank is much larger than the 0.14 Wh/VA consumer-tower ratio assumes. That is the same limitation the layout note already disclosed for rackmount units; the FAQ accuracy answer now states it with these figures instead of a flat plus-or-minus-20% claim it does not meet. (3) The overload guard divided the load by the UPS-TYPE power factor (0.9 or 1.0) to get load VA. That figure describes the UPS output, not the load, so the VA limit bit at 0.9 x the VA rating and returned 0 minutes for loads at or below some units’ rated watts - on the PR2200LCDRT2U every load from 1936W to 1980W, including exactly its rated 1980W, where CyberPower publishes 3 minutes. It now applies a 0.95 LOAD power factor, the conservative end of the 0.95-0.99 this page’s own VA-vs-watts FAQ gives for PFC supplies. The status banner tests the same condition the runtime function enforces, so a 0-minute result can no longer render as yellow ’near capacity’ beside a positive headroom figure, and the runtime table prints 0 rather than ‘<1’ for an overloaded row. (4) The derate coefficients were deliberately NOT refitted; the reasoning is in the code comment and in the sources list.
2026-08-24 — Preset correction from the same evidence class as 2026-08-14: the APC BR1500MS2 preset carried 1500VA / 865W in both the option value (which feeds the watt cutoff and headroom math) and the visible label. Schneider’s live product page titles the unit ‘APC Back-UPS Pro, 1500VA/900W’ (se.com/us/en/product/BR1500MS2/, fetched 2026-08-24). Corrected to 900W. The APC SMT1500 preset (1500VA / 1050W) remains UNVERIFIED and was deliberately left alone: the page checked tonight (SMT1500RM2UC, 1440VA/1000W) is the 2U rackmount SKU, not the tower SMT1500 this preset names, and correcting one SKU from another SKU’s spec sheet is how wrong numbers get shipped. Logged as an open row in the calculator audit TSV.
2026-08-14 — Two of the UPS models this page used to recommend do not exist in CyberPower’s catalog. “CP2200PFCLCD (2200VA / 1320W)” appeared in the preset list, the recommendation cards, the sizing table and all three worked examples; “CP650E” was the sizing table’s 650VA recommendation. CyberPower’s own site search returns 0 results for both, and its PFC Sinewave line tops out at the CP1500PFCLCD. They have been replaced with manufacturer-listed units: the PR2200LCDRT2U (2150VA / 1980W, Waveform: Sine Wave) and the CP850PFCLCD (850VA / 510W, Waveform: Sine Wave). Separately, the CP1500PFCLCD’s watt rating was stated here as 900W; CyberPower publishes 1500VA / 1000W on the very product page this article cites. The preset, the calculator default, the sizing table and every worked-example figure derived from it (percent-of-capacity and headroom) have been recomputed at 1000W. The FAQ’s claim that every listed model outputs pure sine wave was also scoped back: it is verified for the CyberPower units and unconfirmed for the APC BE650G1, whose spec page blocks automated retrieval.
Corrections: Spotted an error or a stale number? Email hello@techfuelhq.com. Confirmed corrections are added to the update log above.