Site planning

AI Server Power, Cooling and Noise: UK Site Guide

Plan a UK site for a GPU or AI server: circuits, kW, electricity, cooling overhead, heat, noise, racks, UPS and pre-delivery checks.

10 minute read Reviewed 25 July 2026 Next review by 25 October 2026

Site system map

Follow every watt through the room

The server is only one part of the installation. Power becomes compute and heat, while the room, rack and operating process determine whether that heat can leave safely.

  1. Supply Circuit, connector, PDU and UPS
  2. Server Measured wall draw under the accepted workload
  3. Air path Clear cool intake and separated hot exhaust
  4. Room Cooling plant, noise boundary and monitoring owner

Check the chain against the final build. A component power figure cannot establish whole-system draw or room suitability.

Worked planning method

Convert load into an energy question

Use measured or conservatively planned inputs. Keep server energy and cooling overhead visible as separate assumptions.

  1. 1.0 kW Illustrative average while the workload is active
  2. 8,760 hours One continuous year before utilisation is applied
  3. 35% Illustrative loaded utilisation, not availability
  4. 3,066 kWh Before cooling, UPS and facility losses

Multiply the result by the organisation’s current contract rate, then add a measured facility overhead. Replace every illustrative input before approval.

A dense GPU server turns electrical energy into useful compute and heat. The room, circuit and operating process are part of the system. Confirm them before procurement, not when the pallet arrives.

Power is a measured range

GPU thermal design power is not whole-system wall power. CPUs, memory, storage, fans and power-supply losses also contribute. Actual draw varies with model, utilisation, power limits and workload.

Use a planning load for the site check, then measure the final build at idle and during the accepted workload. The proposed Opace reference range currently uses roughly 1.0kW to 3.3kW planning loads, subject to exact configuration.

Convert kW into energy

Annual server energy is:

loaded kW × hours per day × days per year × utilisation adjustment

Cooling, UPS and facility losses can be represented by a PUE or simple overhead factor. A 1.5kW server drawing that average for eight hours each working day is very different from a 3.3kW platform operating continuously.

Use the electricity calculator with the organisation’s contract rate. The public model’s 24.14p/kWh input is dated and editable; it is not a universal tariff.

Circuit and connector checks

A qualified person should confirm:

  • available voltage, current and circuit capacity;
  • dedicated-circuit requirement;
  • plug, socket and power-distribution unit;
  • redundant power-feed design where relevant;
  • cable route and safe isolation;
  • UPS capacity and expected runtime;
  • controlled shutdown behaviour.

Do not select electrical equipment solely from a website specification. Final component draw, transient behaviour and local installation conditions matter.

Cooling is not just room temperature

The room must remove the heat the server releases. Front-to-rear rack airflow needs a clear intake and exhaust path. Recirculating hot exhaust can raise GPU and component temperature even when a wall thermostat looks acceptable.

Record intake temperature under load, component temperatures, fan speed and any throttling. Keep vents, filters and rack doors in the maintenance plan.

Where air conditioning is required, include its energy and maintenance in total cost. A small office split unit may not be designed for a concentrated continuous IT load.

Noise can rule out the location

High-density 4U servers move a lot of air. Under load, their fan profile may be unsuitable for a normal shared office. Acoustic cupboards need engineered airflow and fire/electrical review; putting a hot machine in a sealed cabinet is not a solution.

Use a server room, plant space or colocation when human comfort and sustained cooling cannot both be achieved.

Rack and delivery

Confirm rack unit capacity, rail compatibility, usable depth, weight limit, access to the rear and service clearance. Measure doors, corridors, lifts and loading access.

The initial Opace proposition is source configuration followed by customer collection or insured pallet delivery and remote onboarding. Rack installation, electrical work and HVAC are separate unless explicitly quoted.

Monitoring after handover

Track:

  • wall power or PDU power where available;
  • GPU power, utilisation and temperature;
  • CPU and system temperatures;
  • fan and power-supply state;
  • filesystem capacity and health;
  • room intake temperature;
  • hardware errors and unexpected throttling.

Set thresholds that somebody owns. A dashboard without an escalation route is decoration.

Site pre-flight checklist

Do not release procurement until the customer confirms:

  1. physical location and access;
  2. rack space, depth and rails;
  3. electrical capacity and connection;
  4. cooling and exhaust path;
  5. noise acceptance;
  6. network, VLAN, DNS and remote management;
  7. UPS and shutdown policy;
  8. delivery route and lifting plan;
  9. insurance and asset ownership;
  10. named technical and facilities contacts.

If any material item remains unresolved, redesign the location or choose a different delivery model.

Technical context

See the physical and operating boundary

Use these views to connect the guide to the machine, its airflow and its operating environment. Captions state the limits of what each image shows.

Diagram of cool air entering a rack server, heat leaving it and checks for the circuit, room and meter
Diagram of cool air entering a rack server, heat leaving it and checks for the circuit, room and meter
Power, airflow and room conditions form one site-readiness decision rather than three separate afterthoughts. Original explanatory plate. It sets out a decision method, not a measured result.
Power, airflow and room conditions form one site-readiness decision rather than three separate afterthoughts.
Hot-swap GPU server cooling fan module with a pull handle and protective grille
Hot-swap GPU server cooling fan module with a pull handle and protective grille
Supplier component image showing a removable cooling fan module. Service terms and spare-part availability require written confirmation. OEM supplier reference image. Written reuse permission pending.
Supplier component image showing a removable cooling fan module. Service terms and spare-part availability require written confirmation.
Open 4U OEM GPU server chassis showing passive GPUs, cooling fans, processors and memory slots
Open 4U OEM GPU server chassis showing passive GPUs, cooling fans, processors and memory slots
OEM supplier render showing one possible internal layout. Components vary with the ordered build. OEM supplier reference image. Written reuse permission pending.
OEM supplier render showing one possible internal layout. Components vary with the ordered build.
Diagram of an evidence pack containing an asset schedule, burn-in record, health readings, workload test and admin guide
Diagram of an evidence pack containing an asset schedule, burn-in record, health readings, workload test and admin guide
A credible handover records the supplied assets, checks, operating evidence, instructions and unresolved items. Original explanatory plate. It sets out a decision method, not a measured result.
A credible handover records the supplied assets, checks, operating evidence, instructions and unresolved items.

Release gate

Do not book delivery until the site is signed off

A practical pre-flight joins facilities, IT and the supplier around one written record.

  1. Space and route Rack dimensions, weight, doors, lift and service clearance
  2. Qualified check Capacity, connection, isolation, UPS and shutdown
  3. Heat rejection Intake, exhaust, cooling duty and noise acceptance
  4. Named contacts Facilities, network, monitoring and escalation

Any unresolved item changes the design, location or delivery model. A website specification is not a site approval.

Primary sources

Sources are checked at the review date. Platform terms, prices and public guidance can change; verify them at the point of decision.

Use the guidance

Make the next conversation specific.

Describe the workload without uploading confidential material.