Sizing a UPS for a PoE switch is not one calculation. It is three numbers — real AC load, usable watt-hours, and the watt rating rather than the VA rating — plus one check almost nobody makes: what the output looks like when the unit is actually running on battery. Get the first three right and the fourth wrong, and the switch still drops the whole camera group two seconds into the outage.
I watched that happen in a warehouse off Bao'an Avenue last month. Nine minutes of mains outage, a UPS whose runtime label claimed forty. The NVR stayed up and recorded seven minutes of black frames. Nothing in that building was obviously miswired, which is what made it worth writing down.
Nine Minutes of Outage, Two Seconds of Camera
The rack had two UPS units. The small 600 VA one fed the NVR and a desk fan. The big one was supposed to feed the switch. What I found first was the switch on a surge strip, so I said the obvious thing out loud in front of the customer: someone plugged it into the wrong panel.
Then I checked the NVR timestamps and they did not fit. The cameras had gone dark, come back for about ninety seconds, gone dark again, and stayed dark. A switch on a dead strip goes dark once.
It was on the big UPS after all, on a circuit I had not traced. When mains dropped, the unit transferred in about four milliseconds and then shut its output down after two seconds with an overload indication, at roughly 40 percent of rated load. On battery it produced a stepped approximation of a sine wave. The switch supply has an active power-factor-correction front end, which conducts only near the crest of the waveform. On a stepped wave those narrow conduction windows land on a near-vertical edge, peak current triples for a fraction of a cycle, and the UPS reads that as a short and protects itself.
I have been wrong in the other direction too. Years ago I made a client replace a perfectly good modified-sine unit because I had decided waveform was the fault. The real problem was a 9 Ah battery cooked to half capacity by a cabinet sitting at 38 °C. The replacement bought six weeks. Both stories point at the same thing: a UPS is four independent limits, and the weakest one sets the runtime.
VA, Watts, Watt-Hours: Only One Is Capacity
VA is apparent power. Watts is what heats your equipment. The ratio is the power factor, usually 0.6 on consumer models. So a box sold as 1500 VA is often a 900 W box. Load that to 850 W and you are at 94 percent of real capability while the marketing number suggests 43 percent headroom. I now read only the watt rating and treat VA as decoration.
The third number is the one that sets minutes. A 12 V 9 Ah block holds 108 Wh nominally, and that is the only figure that converts to time. VA tells you whether the inverter can carry the load; watt-hours tell you for how long.
And the AC load at the wall is not the PoE budget. A 16-port switch with a 370 W budget does not pull 370 W unless you have loaded all sixteen ports to their class limit. On a camera installation you will measure 130–180 W, because cameras draw what they draw. Sizing a UPS to the PoE budget is the most common way to spend three times what you needed and still get a short runtime. The switch product line is where those budget figures come from, and they are worst-case numbers by design.
What the Switch Actually Pulls From the Wall
The measurement takes ten minutes and it has to be done at night. A fixed camera rated 8 W may sit at 11 W with its IR illuminator on, and a PTZ with a heater can nearly double. Measure at 2 p.m. and you size for a load that never occurs.
- Put a plug-in energy meter between switch and wall and leave it overnight. Take the highest sustained reading, not the average.
- Read per-port PoE wattage from the switch's management page. Add about 0.2 W per port for a 70 m run at 24 AWG — that is copper acting as a heater, and it comes out of the port budget, not the camera's.
- Subtract. If the meter says 165 W and the ports total 136 W, the 29 W difference is conversion loss, fans, PHYs and the management processor. It does not shrink when cameras are few.
- Add the NVR only if it shares the UPS, and check whether its nameplate already includes its PoE output. On an NVR with built-in PoE ports, adding camera watts on top double-counts by a third.
- Add the ONT, the router, and any media converter at the far end of a fibre link. I have seen a well-sized UPS fail its purpose because the converter three rooms away was on a wall wart.
- Multiply by 1.2 for growth and stop. That is your AC load; everything after it is arithmetic.
| Item | Nameplate | Measured at the wall | Note |
|---|---|---|---|
| 16-port PoE switch | 370 W PoE budget | 152 W | 12 cameras, 136 W at the ports, IR on |
| NVR, 4 drives | 120 W | 78 W | Drives idle-spun during the reading |
| Router plus ONT | 36 W combined | 21 W | Needed for remote viewing |
| Monitor, normally off | 28 W | 0 W | Excluded, and it should stay excluded |
| Total to back up | 554 W of labels | 251 W | 1.2 growth factor gives 301 W |
Two hundred and fifty watts against five hundred and fifty-four of labels. That gap is where UPS budgets die, and it is why a 1500 VA unit with two 9 Ah blocks feels disappointing at 300 W.
From Watt-Hours to Minutes
Usable capacity is not nominal capacity. A lead-acid block should not go below about 60 percent depth of discharge if you want more than a hundred cycles out of it, and the inverter loses 10 to 15 percent. So:
Runtime (hours) = battery Wh × 0.6 × 0.88 ÷ load (W)
For the warehouse: 216 Wh × 0.6 × 0.88 ÷ 300 W = 0.38 h, about 23 minutes. The datasheet chart for that class of unit at 300 W says 18 to 22, and the chart is closer, because the formula ignores Peukert's effect — pull energy out of a lead-acid cell faster and you get less of it. Use the formula to sanity-check a vendor claim. Use the chart to plan.
| Battery capacity | 80 W | 165 W | 300 W | 450 W |
|---|---|---|---|---|
| 108 Wh (1 × 12 V 9 Ah) | 43 min | 21 min | 11 min | 8 min |
| 216 Wh (2 × 12 V 9 Ah) | 86 min | 41 min | 23 min | 15 min |
| 432 Wh (4 × 12 V 9 Ah) | 171 min | 83 min | 46 min | 30 min |
Read it as a target check, not a shopping list. Four hours of recording is an external battery string, not a single-box UPS. Ninety seconds until the generator picks up is almost anything.
Which is why I stopped treating runtime as one number. There are two: until the battery is empty, and until the first device misbehaves. The second arrives earlier, because a switch that reboots drops every PD at once, and thirty seconds of camera boot time is thirty seconds of nothing recorded. If only some cameras must survive, the priority and shedding logic is worth more than another battery block.
The Waveform Problem Nobody Checks
Every switch supply above 75 W has an active PFC stage, because harmonic limits require it. Feed that stage a stepped waveform and the current spikes land on the fast edges. The UPS measures output current, sees a fault, and does its job. The failure looks like a UPS fault and is really an impedance mismatch between two decent products. Symptoms I now recognise on sight:
- Overload the instant it transfers, at well under half rated load.
- Fine on mains, fails only on battery — which makes people blame the battery.
- A smaller, older switch on the same unit runs fine, because a 40 W supply has a proportionally gentler PFC stage.
- The unit buzzes audibly on battery, louder than on mains.
Transfer time is the other half. Standby units switch in 6 to 10 ms, line-interactive in 2 to 4, double-conversion in zero. A switch supply holds its rails for roughly 10 to 20 ms at full load, and that figure comes from the bulk capacitors — so it shrinks as they age, the same wear I described in the capacitor life article. A four-year-old supply with dried electrolytics may hold for 8 ms. Behind a 10 ms standby unit, that switch reboots on every transfer, which is exactly the site where the customer says it works fine except during outages.
My rule now: above 150 W with an active PFC supply, specify pure sine output. Not because the brochure says so, but because I have watched a 600 VA modified-sine unit refuse a 240 W PoE switch, and then watched a 1000 VA pure sine unit carry the same switch for 26 minutes on the same battery capacity. The second one cost more. It also worked.
Batteries Die in Warm Cabinets, Not on Schedule
A VRLA block is rated three to five years at 25 °C. Every 10 °C above that halves the life — the same Arrhenius relationship as electrolytic capacitors, for the same reason. A UPS in a wall cabinet that reaches 35 °C in summer is a two-year battery. In an outdoor cabinet in Guangdong with the sun on the door from two in the afternoon, I have pulled blocks that were finished at eighteen months.
Cold is a different problem. Capacity falls to roughly 80 percent at 0 °C and 65 percent near −15 °C, so a unit sized in July comes up short in a cold snap. Outdoors, size for the winter number and expect summer to eat the battery's life instead.
- Write the commissioning runtime on a label inside the door: date, load in watts, minutes measured.
- Once a year, open the input breaker with the real load connected and time the run to low-battery alarm. Not the self-test — that runs a few seconds at whatever load is present and passes on a battery that is 60 percent gone.
- Replace below 80 percent of the commissioned figure, not when the unit beeps. The beep means the battery is already dead.
- Log cabinet temperature at the same time. Above 30 °C, shorten the interval instead of arguing about it.
- After any replacement, re-run the transfer test with the switch connected. New batteries do not fix a waveform problem, and you want to know that before the next outage.
The standard people cite for this is IEEE 1188, on VRLA testing and replacement intervals. Worth reading once. What it will not tell you is that inrush on restore is the second event that kills a marginal UPS — twelve cameras and their illuminators starting within a second of each other, which is the startup surge problem arriving on the UPS output instead of the PSU output. If the unit survives the transfer and dies on restore, that is what happened.
Questions I Get Asked
Can I run a PoE switch on a 600 VA UPS?
Usually, if the load is genuinely small — eight cameras at 60 W total sits well inside 600 VA. Two conditions: read the watt rating (360 W at 0.6 power factor) rather than the VA, and check the waveform above 150 W of supply. That combination has failed twice on sites I have visited.
Should the NVR and the switch share one UPS?
Yes, and this is the mistake I see most. Switch on a surge strip, NVR on the UPS, and an outage gives you a recorder watching dead cameras. The whole chain — switch, NVR, router or ONT — belongs on battery-backed outlets. Watch the units that split outlets into surge-only and surge-plus-battery; the second group is usually half the panel and people plug into whichever socket is free.
How much runtime should I specify?
Ask what the system has to do, not how long the outage lasts. Ride through a generator start: two minutes is generous. Bridge a recloser: fifteen to thirty. Keep recording for hours: you are specifying a battery bank, not a UPS. "Thirty minutes" with no stated purpose satisfies the purchase order, not the site.
Does a bigger UPS give longer runtime?
Only if it has a bigger battery. Many larger units carry the same two 9 Ah blocks with a beefier inverter — more load, same minutes. Compare watt-hours and ask for the runtime chart at your actual load. A 3000 VA unit with 216 Wh runs your 300 W rack exactly as long as a 1000 VA unit with the same blocks, for three times the money.
What about a 48 V DC plant instead of a UPS?
Real option for industrial sites, and it removes both the inverter loss and the waveform question — many DIN-rail switches accept 48–57 VDC directly, which is what the industrial range is built around. Two cautions. A 48 V lead-acid string sags to about 44 V near end of discharge, so a PSE that cannot hold its output above the PD minimum starts dropping ports while the battery still has charge. And the charger needs three-stage control with low-voltage disconnect, not a float supply, or the string is cooked inside a year. In an office rack I would not argue for anything but AC and a pure sine UPS.
If you are sizing one of these next week, bring three numbers to the table: measured AC load at night, battery watt-hours, and output waveform. Everything else on the datasheet is negotiable. I would like to hear from anyone who has measured the real transfer behaviour of a small PoE switch behind a line-interactive unit — my sample is a dozen units on one bench in Shenzhen, and a dozen is not a dataset. The meters and the outage logs have never lied to me; half the time they have told me I was wrong about which part failed.