How to size a UPS for a home lab: VA, watts and runtime

To size a UPS for a home lab, measure what your gear draws with a plug-in meter, then pick a unit whose watt rating is at least twice that figure and whose VA rating sits well above the VA the meter shows. Read the runtime off the maker's chart at your measured load, not at the unit's rating. If anything on the list has a computer-style power supply with active PFC (a NAS with an internal supply, a desktop, a tower server), buy pure sine wave output. A closet with a modem, a router, a switch and a Raspberry Pi needs no more than a 600 to 850 VA desktop unit. Add a hypervisor box and you are into 1500 VA, pure sine territory.

# Affiliate links: we may earn a commission. As an Amazon Associate I earn from qualifying purchases. We also earn from qualifying purchases on eBay.

Three questions before you look at a single model

1. What has to stay up, and what only has to shut down cleanly? The modem, router and switch carry the alerts and shutdown messages, so they want the longest runtime. A NAS or a Proxmox host mostly needs a few calm minutes to stop.
2. How much power does it draw? Measured at the wall. A 65 W adapter or a 550 W supply only tells you the ceiling.
3. How long must it run on battery? Minutes are a UPS job; hours are not (see the last section).

What to look for

VA, watts and power factor

VA is volts times amps, each measured on its own: the apparent power the UPS's electronics carry. Watts are the real power, the part that does work and ends up as heat. Power factor is watts divided by VA, somewhere between 0 and 1.

A UPS has a limit on both, and your load must fit under each. The APC BE600M1 (Amazon ›) is rated 600 VA and 330 W, so its watt limit is only 55 % of its VA figure. Shoppers compare the big VA number; the watt number is the one that usually stops you.

A computer supply with active PFC runs a power factor close to 1, so watts bind first. The small adapters feeding modems and switches usually have no PFC and a power factor well below 0.9, so a closet of wall bricks uses more VA than its watts suggest.

Measuring the real load

Put a plug-in energy meter between the wall and the power strip feeding everything you plan to protect; most show watts, VA and power factor. Leave it a day, because the load moves: PoE cameras switch on infrared at night, backups spin up disks at 02:00, everything pulls more at boot. Write down the highest steady figure, not the spike at power-on.

Once a UPS is in, models with an LCD show the load and an estimated runtime (the BE600M1 has LED bar graphs), and NUT reports ups.load as a percentage of capacity. Those readings are coarse at light loads: size with the meter, and use the UPS for the trend.

Runtime is not linear

CyberPower's published figures for the CP1500PFCLCD (Amazon ›) are about 10 minutes at half load (500 W) and about 2.5 minutes at full load (1000 W). Double the load and runtime drops to a quarter. In energy terms the same battery delivers 500 x 10 / 60 = 83 Wh at half load and only 1000 x 2.5 / 60 = 42 Wh at full load.

Lead-acid gives up less capacity the faster you drain it, the inverter loses some energy, and the UPS stops at its low-battery cutoff before the battery is flat. It works in your favor too: as the load falls, runtime grows faster than the load shrinks, so shutting a NAS down early buys the network gear real time.

So read the chart at your measured watts. APC notes that its runtime graphs come from new, fully charged batteries. Yours will not stay new, so plan with a margin. To compare units before you have their charts, the UPS runtime calculator estimates runtime from the battery inside each one, with the range real charts tend to land in.

Pure sine or simulated sine

On battery, the inverter makes its own waveform. Pure sine units (the CP1500PFCLCD and the BR1500MS2 (Amazon ›) below) make a clean sine wave. The standby units make what APC calls a stepped approximation to a sine wave and CyberPower calls simulated sine wave.

Wall adapters, routers, switches and a Pi on its USB-C supply generally cope with that for the minutes a UPS runs. The trouble is active PFC supplies inside desktops, servers and some NAS units: some switch off or reboot when they see the stepped wave at transfer, others carry on, and you find out which during a real outage. Pure sine removes the question.

Standby or line-interactive

A standby UPS feeds your load straight from the mains through its surge filter and switches to the inverter when the power fails. APC's datasheets give the BE600M1 and BE850G2 a transfer time of 6 ms typical, 10 ms maximum; CyberPower quotes 4 ms for the EC850LCD. A line-interactive unit adds an autotransformer (AVR) that boosts low voltage and trims high voltage without touching the battery.

If your lights dim when the air conditioner starts, a standby unit goes to battery for every sag, wearing the battery. A line-interactive unit rides those sags on its transformer.

Count the outlets with a battery behind them

Every unit here has surge-only outlets that go dark in an outage. From the makers' spec sheets:

The classic mistake is the modem in a surge-only socket: the router stays up and the internet still goes down. Wall bricks block neighboring sockets, so count usable outlets. Keep laser printers and heaters off the battery side.

A worked example: one closet, then a bigger lab

A typical closet, measured with the meter on the power strip:

cable modem                       11 W
router                             9 W
8-port PoE switch + 2 APs         38 W
Raspberry Pi 5 monitor             7 W
2-bay NAS (external 12 V brick)   31 W
total on the meter                96 W   140 VA
power factor                      96 / 140 = 0.69

Capacity: twice 96 W is 192 W, and 140 VA is far under every VA rating, so all five units pass. On the smallest, 96 / 330 = 29 % of the watt rating.

Waveform: nothing here has an internal PFC supply, so a stepped wave is fine.

Outlets: five devices need five battery outlets. The BE600M1 has exactly five, nothing spare for next year. The 850 VA units have six.

Runtime: ride out 5 minutes with everything on, then NUT shuts the NAS down, which takes 2 minutes, then 10 more minutes for the network gear and the Pi to send alerts. That is 5 + 2 + 10 = 17 minutes. I count all of it at the full 96 W, which overstates the need, since the load drops to 65 W once the NAS is off. Add 25 % for an aging battery: 17 x 1.25 = 21.25, so I look for at least 22 minutes at about 100 W on the maker's chart. The 850 VA units are the comfortable fit.

Now the bigger lab: add a Proxmox mini tower with an active PFC supply that the meter shows at 160 W when busy. The load becomes 96 + 160 = 256 W; twice that is 512 W, which rules out the 450 W and 510 W units, and the PFC supply wants pure sine anyway. That leaves the 1500 VA class: 256 / 1000 = 26 % of the CP1500PFCLCD's watt rating, 256 / 900 = 28 % of the BR1500MS2's.

The Rack list

These five cover the sizes above, from a modem-and-router closet to a small rack with a hypervisor, and all have a USB data port. The list comes from manufacturer specifications and published reviews; none of these units has run on my bench.

Setup and monitoring notes

Connect the data cable that came in the box; a USB charging socket, where a model has one, carries no data. Most USB units from both makers present themselves as HID power devices, which the usbhid-ups driver in Network UPS Tools reads; check the NUT hardware list for your model. The full setup, from upsd to a tested shutdown order, is in UPS monitoring with NUT. In the example closet, the Pi from the weekend Raspberry Pi monitoring build makes the natural NUT primary, since it stays up to the end.

Three commands for day one: read the status, list what the UPS accepts, and run a quick self-test if it offers one:

upsc rackups@localhost ups.status
upscmd -l rackups@localhost
upscmd -u admin rackups@localhost test.battery.start.quick

Graph ups.load, battery.runtime and input.voltage next to your other graphs (the field manual's page on what you can monitor shows how they fit), and alert on battery, low battery, lost communication and replace battery.

Battery replacement

The battery wears out first. APC gives the BR1500MS2's battery a design life of 3 to 5 years and the BE850G2's 2 to 5. Heat shortens it: APC's rule is that every 8.3 °C (15 °F) above 25 °C (77 °F) halves battery life, so a UPS in a warm closet ages faster than one in a cool basement.

Time to replace it when the replace-battery light or the RB flag in ups.status shows, a self-test fails, or the runtime graph at a steady load slides month after month. Order the part number on the datasheet (RBC17 for the BE850G2, APCRBC163 for the BR1500MS2), let the new battery charge before you trust it, and take the old one to a recycler. Recharging is slow, too: APC lists 8 hours for the BE850G2 and 16 for the BR1500MS2, so a second outage the same evening gets less runtime than the first.

When to spend more

Spend more when the requirement changes. A UPS with a network management card speaks SNMP, useful once several machines share it; snmpwalk against the card is the first test.

For hours rather than minutes, a desktop UPS is the wrong tool. A portable power station such as EcoFlow or Jackery holds far more energy, often in lithium iron phosphate cells that outlast lead-acid. The trade-offs: makers quote switchover times for their backup modes from under 10 ms on some newer models to around 30 ms on others, against 4 to 10 ms on the UPS spec sheets above, and a long switchover can reboot a PC. Many offer no data link NUT can read, so nothing tells your machines to shut down. Put the power station in front of the network gear and keep a UPS with NUT in front of anything with a disk.

Before you buy, run the list once more: measured watts and VA, twice the watts in rating, pure sine for PFC supplies, one battery outlet per device plus a spare, and a runtime figure from the chart with a quarter added for age.