Loads: racks, cooling, UPS
A server room starts by totalling three loads: IT racks, cooling and the UPS itself with efficiency and battery-charge current. Racks draw a dense, near-24/7 load — take it metered or nameplate plus growth margin. Cooling often eats a third or more of the supply: calculate it separately on independent groups.
| Consumer | Character | Supply requirement |
|---|---|---|
| Racks (PDUs) | Dense 24/7 load | Redundant groups, row metering |
| Cooling units | Compressor starts | Separate groups, never off UPS |
| UPS | Battery charging, efficiency | Incomer with margin, bypass |
| Light, access, fire alarm | Small but critical | Own breakers |
The “racks only” mistake is unforgivable: in summer, cooling plus post-dip UPS charging easily consumes the whole incomer margin.
Clamp-meter existing circuits when expanding a room rather than building from scratch: PSU nameplates usually overstate real draw, and measurements reveal the honest margin. Allow space and capacity for growth — at least one or two extra racks and a second cooling unit. Spare panel ways today cost less than a panel replacement next year when the new cluster arrives.
Redundancy and bypass without magic
Server-room redundancy is built from simple techniques, not magic. Baseline: a UPS with external bypass — servers ride through dips and short outages, and the bypass lets you service or replace the UPS without dropping racks. Next level: two supply sections in the panel, so servicing one never blacks out everything. Dual-PSU equipment feeds from different sections.
State autonomy honestly: batteries cover the minutes needed for graceful shutdown or reserve start-up, not “hours for the whole office”. How much time your room needs is the owner's call based on service criticality. No design should promise absolute, outage-free operation: UPS and redundancy cut risk but never replace maintenance schedules and battery testing.
Groups and PDUs
Wire the room so one faulty element never drops its neighbours. Cooling sits on dedicated groups from the power panel, bypassing the UPS: running compressors off batteries is pointless and expensive. Each rack or row gets its own circuits from the server-room panel via metered PDUs — so you see who eats capacity before a breaker trips.
- Racks — redundant groups from different sections for dual-PSU kit.
- Cooling — dedicated groups straight from the panel, never off UPS.
- Lighting, access control, fire alarm — their own breakers.
- Spare panel ways for growth: a new rack must not mean a new panel.
Verify protection selectivity via time-current characteristics: a short in one rack must trip its own device, not the incomer. Short-circuit and selectivity calculations are preliminary materials the engineer approves.
Metered PDUs turn operations into numbers: you see which rack approaches its group limit, and new loads land where spare capacity exists instead of “where the cable reached”. Log readings regularly — a year later they become the main argument for expanding the room. Labelling is mandatory: every circuit, breaker and PDU is marked so night-shift staff find the right group without calling the engineer.
Monitoring, metering and access
A server room without metering breeds arguments about “who ate the capacity”. Meter the room incomer and, ideally, rack groups: meters or metered PDUs show real consumption and the moment to buy more power or cooling. Restrict physical access: lock the server-room panel and keep unrelated consumers off its groups.
On monitoring, plainly: the design provides metering points and records, while remote collection and dispatch systems (SCADA/IoT) are not a GorkyCAD feature — provide them separately if needed. The project records the ops routine: who checks UPS batteries, contact tightness and meter logs, and how often.
Documents and ops rules
The server-room set includes the rack and route layout, the single-line with UPS and bypass, the panel schedule, the cable schedule, the bill of materials and the estimate — one PDF package from one model. The single-line generates automatically with standard device symbols (IEC 60617 in EN practice). Calculations run in a deterministic standards-based engine, not a language model — repeatable and verifiable.
Attach the ops routine: bypass switching procedure, battery-check intervals, emergency actions and responsible persons. Until the engineer signs, the whole set counts as preliminary materials. Objects subject to mandatory expert review pass it before installation.
After installation, run acceptance tests: insulation resistance, protective-conductor continuity, trip checks and phasing of redundant lines. File the records with the set — without them the maintenance routine has nothing to anchor to: the first measurements become the baseline for every later check.
Classic server-room power mistakes
- Cooling on UPS: batteries die in minutes, and compressors never survive long outages anyway.
- One group for all racks: a short in one PDU drops the whole row.
- No bypass: a battery swap becomes a service outage.
- UPS charge current ignored: post-dip charging eats the remaining incomer.
- No group metering: growth is discovered via a tripped breaker.
- A panel with no spare ways: every new rack means rebuilding the panel.
Automated checks — 25 checks to IEC 60364 — flag some violations before installation, but decisions and signature stay with the engineer.
Designing a server room in GorkyCAD
Draw the room in the browser: one model links plan, single-line with UPS and bypass, the server-room panel, specification and estimate. Topology builds top-down — main board to the server-room distribution panel — and the auto-panel lays devices on DIN rails. The AI assistant helps with commands and checks, but calculations run in the deterministic engine, never in the AI. Works offline as a PWA.
Start free in the editor, and unlock the full set with calculations and exports on Pro at €40/mo. Open the editor and lay out the server room →
FAQ
How many circuits per rack?
Common practice is two independent circuits from different panel sections for dual-PSU equipment, plus current metering on each via metered PDUs. One circuit suits non-critical racks, but then any group maintenance means downtime. The engineer sets the exact count from equipment class and the redundancy scheme.
One UPS for the room or per group?
A small room usually needs a single UPS with external bypass for the racks, with cooling and light fed past it. Per-group UPS units suit rooms where racks have different owners or autonomy demands. Always include a bypass: without it, UPS servicing stops services.
Cooling on its own circuits?
Yes — always dedicated groups straight from the panel, never off UPS: compressors draw starting currents and waste batteries pointlessly. Cooling is its own load-calculation line: in summer it plus post-dip battery charging easily consumes the whole incomer margin. Allow for a second cooling unit too.
Does a small room need a generator?
Usually not: a small room is fine with a UPS sized for graceful shutdown plus a written long-outage procedure. A standby generator suits services that must never stop — the owner's call on criticality, not a mandatory element of every server room. The design honestly states autonomy time instead of promising “no outages”.
Related articles
- Emergency power: sources and reserve schemes
- Generator selection: sizing and connection
- Selectivity explained in plain terms
- Switchboard checklist before handover
Design the server room: racks, UPS and panel from one model
Draw the server room, build the UPS-and-bypass scheme and export the PDF set with ops rules. Start free in the browser — Pro €40/mo with full calculations.
Open GorkyCAD