UPS Sizing for a Mini PC Homelab: Runtime, VA Math, and What I Actually Run

A mini-PC homelab pulls so little power that picking a UPS feels like it should be trivial. It isn’t — not because the load is hard to support, but because most UPS marketing is written for desktop towers and small office racks. Sized that way, you end up buying twice the unit you need, with worse runtime, for more money. This guide walks through the actual numbers for a 10–30W homelab, the trap with cheap line-interactive units, and the specific models that earn their slot.

What you’re actually protecting against

A homelab UPS does three useful things, in this order:

1. Survives short outages (under 30 seconds) so nothing reboots when the lights flicker. This is 90% of the value.
2. Gives you enough runtime to shut down cleanly on a longer outage so you don’t corrupt a database or filesystem mid-write.
3. Cleans up voltage sags and surges that would otherwise stress the PSU on your mini PC.

What it does not need to do, for most people: keep services up through a multi-hour outage. That’s a generator or a battery wall, not a $90 desktop UPS. Size for “ride through the flicker, then shut down gracefully.” Anything more is over-spec.

The two numbers that actually matter

UPS units are advertised in VA (volt-amps), but for a homelab what you actually care about is:

• Runtime at your real load — minutes the battery will hold up your kit at the wattage you draw, not the headline runtime at the unit’s rated load.
• Whether it has a USB data port that something like NUT or apcupsd can talk to, so your mini PC can trigger a clean shutdown automatically.

A 600VA / 360W UPS holding up a 200W desktop will give you maybe 4–5 minutes. The same unit holding up a 20W mini PC plus a 5W switch will give you 30–60 minutes, because runtime scales roughly inversely with load until you hit very low draws where inverter overhead dominates. That’s the whole game for a low-power homelab: the smallest “real” UPS on the market is hugely overspec’d for your load, which is exactly what gets you long runtimes for cheap.

If you don’t know your real load, measure it. A cheap Kill-A-Watt or a smart plug with energy reporting will give you a watt figure in two minutes. My measuring mini PC idle wattage post walks through the exact method and the common mistakes.

The trap with cheap line-interactive UPSes and modern PSUs

This is the one thing nobody warns you about. Most sub-$100 UPSes are line-interactive with a stepped-approximated (sometimes called “simulated”) sine wave output when running on battery. On a modern PC PSU — especially the small efficient ones in N100 boxes that use active power factor correction (APFC) — that stepped waveform can cause the PSU to drop out or behave erratically the moment the unit switches to battery. The exact symptom is the lights flicker, the UPS clicks, and your mini PC reboots anyway.

The mitigation: for any homelab where you care about clean APFC behavior, buy a true pure sine wave UPS. That’s the single most important spec in this category. It typically adds $30–$60 over the cheapest line-interactive units but it’s the difference between a UPS that works and one that doesn’t.

The good news: at homelab loads, the cheapest pure sine wave units are plenty. You don’t need 1500VA.

Sizing math for a real homelab load

Let’s run through a typical small setup:

Device	Watts
N100 mini PC (Proxmox + Docker LXC)	12W
8-port unmanaged gigabit switch	4W
Cable modem	6W
Wi-Fi router	8W
Total	30W

A 600VA / ~360W pure sine wave UPS at 30W of load will hold up that stack for roughly 30–60 minutes, depending on battery health and unit efficiency. Vendor runtime charts are reliable enough for planning at these very low loads — check the spec sheet on the model you’re considering and find the runtime line at the watt figure nearest yours. APC and CyberPower both publish these on their product pages (linked in sources).

For a slightly larger stack — say an N305 with 4–6 spinning drives, a managed switch, and a couple of PoE access points — you’ll be in the 60–120W range and want an 850VA–1000VA pure sine unit. Same model families, one tier up.

Once you have your watt figure, the rule of thumb I use:

• Under 50W total load: 600VA pure sine wave UPS is plenty. ~30 minutes runtime, $90–$130 new.
• 50–100W: 850VA / 510W class. ~20–30 minutes runtime, $120–$170 new.
• 100–200W: 1000–1500VA class. ~20–30 minutes, $180–$280 new.

Note the runtimes converge at 20–30 minutes regardless of load tier — that’s because the vendors size the battery to hit a target runtime at the unit’s rated load, not yours. At your sub-rated load you’re getting a bonus, but past about 30 minutes the unit is already idling on inverter overhead, so doubling the VA doesn’t double your runtime in this class.

Models worth buying for a low-power homelab

These three sit in the “buy it, plug it in, done” category for the under-50W mini-PC stack. All are pure sine wave on battery, have a USB data port, and are supported by NUT or apcupsd.

• APC Back-UPS Pro BR700G / BR1000MS. APC’s pure sine consumer line. Well-supported by apcupsd, replacement battery is a standard sealed lead acid you can swap yourself at the 3-year mark.
• CyberPower CP1000PFCLCD. CyberPower’s PFC-compatible Sinewave line is one of the cheapest ways to get a real pure sine wave UPS. The CP1000PFCLCD is the popular homelab pick because it’s overspec’d for low loads, giving you very long runtime. The smaller CP685AVR is line-interactive (not pure sine wave) — fine for a router/switch only, skip it for the mini PC.
• Eaton 5S700LCD / 5S1500LCD. Less common in US consumer channels but excellent build, true pure sine, and very well-supported in NUT.

What to skip for the mini PC line: the rock-bottom $50–$70 line-interactive units (APC BE600M1, BE850M2 and similar) — fine for protecting a router and switch, but the stepped waveform on battery is a real risk for an APFC mini-PC PSU. Save them for the network closet, not the homelab box.

Wire it up so it actually triggers a shutdown

A UPS that doesn’t tell the server it’s on battery is a delaying device, not a shutdown plan. Two pieces:

1. Plug the USB data cable into the mini PC.
2. Install NUT or apcupsd, configure it to monitor the UPS, and set the policy: shut down after the battery has been on for N minutes (I use 5), or when remaining capacity drops below 30%, whichever comes first.

NUT’s hardware compatibility list ↗ is the canonical reference for what works — both APC and CyberPower units listed above are well-supported. For Proxmox, NUT is the cleaner choice because you can run a upsd on the host and have your LXC guests subscribe to shutdown events. For a bare-metal Debian + Docker setup, plain apcupsd is simpler. My Proxmox on a mini PC setup guide covers the LXC-vs-VM split that informs which approach to take.

Test the shutdown trigger by pulling the UPS power cable and watching what happens. Do this once when you set it up, and again after any major OS update. An untested shutdown plan is a guess.

The runtime question people get wrong

The right question isn’t “how long will my UPS last” but “how long do I need it to be up?” If the goal is riding through flickers and grid switching events, under a minute is enough and any UPS in this class works. If it’s keeping services up through a multi-hour outage, you’re in generator or battery-wall territory — wrong tool. Every UPS in the buying tiers above gives you enough margin to shut down cleanly with an automated trigger, which is the actual success criterion.

For the broader power-cost framing on whether long runtimes are worth chasing on your stack, see my mini PC power consumption cost math.

Bottom line

• Pure sine wave on battery is non-negotiable for a modern mini PC PSU. Skip line-interactive units for the homelab box itself.
• 600VA is plenty for a sub-50W stack. You’re not undersizing; you’re getting a long runtime by being well under the unit’s rated load.
• Plug in the USB cable and configure NUT or apcupsd. A UPS that can’t trigger a clean shutdown is doing half its job.
• Buy from APC, CyberPower (PFC line), or Eaton. All three play well with the standard Linux UPS daemons and have predictable battery replacement at the 3-year mark.

The whole UPS for a mini-PC homelab costs less than a single round-trip to fix a corrupted Postgres database, and at 10–30W of load it gives you more runtime than the unit’s headline number would suggest. Buy the right tier, wire the data cable, test the shutdown trigger once, and forget about it for three years until the battery gets tired.