Contact
Power, UPS & load-shedding

Sizing a UPS for Load-Shedding

Power, UPS & load-shedding11 min read

By Humphrey Theodore K. Ng’ambi

Updated 13 September 2026

South African load-shedding turns UPS sizing from a tick-box into a real engineering decision. The grid does not just switch off cleanly for a couple of hours; the ramps up and down bring deep voltage sags, and many sites bridge the gap with a generator or inverter whose output is not as clean as the mains. Get the sizing wrong and you either burn through batteries and cash on protection you do not need, or you buy a UPS that drops the load at exactly the wrong moment. This decoder covers the three things that actually decide it — topology, VA versus watts, and runtime — with the South African feed firmly in mind.

We deliberately quote no rand figure here: runtime, battery count and price all move with your exact load and the model you choose. Use Server Hub's load-shedding runtime calculator to put numbers to your own rack.

The load-shedding problem, in power terms

Eskom's stages each shed roughly 1000 MW — Stage 1 about 1000 MW, up to Stage 8 about 8000 MW. For a given site the practical effect is that outage slots are typically two hours at Stages 1–4 and four hours at Stages 5–8, scheduled several times a day; at the highest stages a site can be off for the better part of half a day, and a short switching allowance is added around each slot (Eskom load-shedding stages; Eskom schedule interpretation).

Two consequences matter for UPS choice. First, the outages are long — hours, not the seconds-to-minutes a UPS is built to bridge on its internal batteries. Second, the power around each outage is rough: sags as demand is shed and restored, and, on sites that run a generator or inverter through the slot, an AC feed whose voltage and frequency are less stable than the utility's. That combination — long outages and a dirty feed — is what pushes some sites from line-interactive to online double-conversion.

VA vs watts (and power factor)

Every UPS is rated in both VA and watts, and they are not the same number:

  • Watts (W) — the real power your equipment consumes.
  • Volt-amps (VA) — the apparent power, the voltage multiplied by the current drawn.
  • Power factor (PF) — the ratio between them: W = VA × PF, and VA = W ÷ PF. A power factor of 0.8 (80%) means a 1000 VA UPS supplies 800 W.

The rule that keeps you out of trouble: your load must sit under both the UPS's VA rating and its watt rating, because it will hit one of them first. On many current UPS the two figures are close; on older or entry-level units the watt figure can be well below the VA figure, so always read both off the datasheet and size to the tighter one. Then add 20–25% headroom above your measured load to cover start-up surges and a little growth (Schneider Electric — UPS buying guide).

The three topologies

UPS designs differ in how the inverter relates to the load. Three matter in practice.

StandbyLine-interactiveOnline double-conversion
Inverter under normal powerOff until power failsConnected, lightly loaded (typically under 10% of rating)Always carrying the full load
Transfer to battery on outageBrief switchBrief switch (server supplies ride through it)None — output is continuous
Voltage regulation (AVR)MinimalYes (tap-changing transformer)Yes (output fully re-generated)
Output frequencyFollows the inputFollows the input, within a windowRegulated; can even convert 50↔60 Hz
Typical power band0–0.5 kVA0.5–5 kVA5 kVA and up (common above 10 kVA)
Typical efficiencyVery high~96–98%~85–92%
Typical useDesktops, till pointsRack and distributed servers — the IT workhorseMission-critical loads, poor/unstable power, generator feeds

Standby simply switches to battery when the mains fails — fine for a desktop, not for a server rack. Line-interactive keeps its inverter connected and adds a tap-changing transformer (automatic voltage regulation, AVR) that corrects moderate over- and under-voltage without going to battery, which is why it is the dominant, efficient choice for rack servers. Online double-conversion converts mains to DC and back to AC continuously, so the load is always fed from the inverter — there is no transfer at all when the mains drops, and the output can be held at a fixed, clean frequency regardless of the input (APC — the different types of UPS systems).

When a server actually needs online double-conversion

The honest answer is: less often than it is sold. Modern server power supplies (switch-mode supplies) are built to ride through brief interruptions — the international UPS standard (IEC 62040-3) treats a very short output gap of a few milliseconds as acceptable, and a typical server supply rides through an interruption on the order of tens of milliseconds. So a good line-interactive UPS's brief transfer to battery does not disturb a rack server. For most cabinets on a reasonable municipal feed, line-interactive is the correct, cheaper, cooler-running answer (APC/Schneider — technical comparison of UPS topologies).

Online double-conversion earns its premium where the feed is the problem — which is exactly the South African edge case:

  • Highly distorted or wildly varying input. Where the AC is dirty or swings hard, an online UPS goes to battery less often, because it regenerates the output rather than transferring. That preserves battery capacity for the actual outage and lengthens battery service life — a real saving where the grid is rough day after day.
  • Generator or inverter feeds with unstable frequency. Only online can hold a fixed output frequency and isolate the load from a wandering source; a line-interactive UPS passes the input frequency through.
  • Loads that need power-factor correction the equipment does not provide, or that need frequency conversion.

Rule of thumb for a South African site: line-interactive for a stable municipal feed; online double-conversion where the feed is dirty, sags hard on every load-shedding ramp, or is generator/inverter-backed. The trade-off is efficiency — online typically runs at ~85–92% against ~96–98% for line-interactive, so it draws more, runs hotter and costs more to operate.

Runtime: why it is not linear, and how to size it

Runtime comes from the batteries, not from the VA rating — a bigger-VA UPS with the same battery does not run longer. And runtime is non-linear: a lightly loaded UPS runs far longer than its headline full-load figure, and as you add load the runtime falls away faster than the load rises. Never estimate runtime by dividing linearly. Read the manufacturer's published runtime-versus-load chart for the exact model.

The load-shedding reality check: a standard internal-battery UPS is designed to bridge seconds to minutes — long enough to ride a blip or shut a server down gracefully. It is not built to carry a rack through a two- to four-hour slot. To span a whole slot you need one of:

  • Extended battery packs sized from the runtime chart (battery strings, not a bigger inverter), or
  • A generator or large inverter carrying the load, with the UPS conditioning its output and covering the changeover.

Sizing a UPS to run a rack for four hours on internal batteries alone is rarely practical or economic. Decide your target first — ride-through plus graceful shutdown, or genuine hours across a slot — then use Server Hub's load-shedding runtime calculator to turn your watt load and target time into a battery/runtime requirement.

A sizing method

  1. List the load. Every device and its real running watts (nameplate is a ceiling, not the actual draw — use measured or typical figures where you can).
  2. Total and add headroom. Sum the watts and add 20–25%.
  3. Check both ratings. If the UPS is rated in VA, convert (VA = W ÷ PF) and confirm the load fits under both the watt and VA ratings.
  4. Pick the topology. Line-interactive for a stable feed; online double-conversion for a dirty/unstable feed or a generator-backed site.
  5. Set the runtime target and read the runtime chart or the calculator; add extended battery packs or a generator to reach a full slot.
  6. Choose the form factor. Tower for a single server or desk; rackmount (2U–4U) to live in the cabinet with the kit it protects; add a network management card for remote monitoring and safe automatic shutdown during a long outage.

Choosing a vendor and form factor

The major names all span the topologies, so match the topology to your feed and the runtime to your slot, then pick the range:

  • APC by Schneider Electric — Back-UPS (standby, desktop), Smart-UPS (line-interactive, the rack workhorse), and Smart-UPS On-Line / Symmetra (online double-conversion).
  • Eaton — line-interactive (e.g. 5P/5PX) and online double-conversion (e.g. 9PX/9SX) ranges for tower and rack.
  • Vertiv (Liebert) — line-interactive and online ranges, tower and rack.
  • CyberPower — value line-interactive and online ranges.

For a rack, favour a rackmount unit that sits with the equipment it protects and takes external battery packs, plus a management card so a long slot triggers an orderly shutdown rather than a hard drop. Server Hub carries both new and certified-refurbished units — see the catalogue for current models.

The Server Hub briefing

South African IT hardware news, once a week.

What’s new, what it costs in rand, and what it means for the kit you run — servers and storage, networking, backup power, surveillance and print. Every claim checked against a named source.

One email a week. No third-party sharing, and unsubscribe from any issue.