Sensible Heat Load Calculation for Server Rooms: A Practical Guide to CRAC Sizing

Sensible heat load calculation is where data centre cooling design either succeeds or fails. Undersize the cooling plant and you risk thermal runaway during peak load. Oversize it and you pay for capacity that never runs efficiently, often at the cost of humidity control. Neither outcome is acceptable in a production environment.

This guide walks through each heat load input category, explains how to treat nameplate ratings versus actual power draw, and covers sensible heat ratio (SHR) in enough depth to inform CRAC unit selection.

What Sensible Heat Load Actually Means

Heat load in a data centre is almost entirely sensible, meaning it raises air temperature without adding moisture. This distinguishes it from latent heat, which changes the moisture content of air without changing its dry-bulb temperature. Understanding the split matters because CRAC units are rated for total cooling capacity, but their ability to handle sensible versus latent load varies by design.

Sensible heat load (Q_s) is expressed in kilowatts or BTU/hr. The total load for a server room is the sum of:

• IT equipment heat dissipation
• Lighting within the space
• Building envelope gains (conduction and solar)
• Human occupancy
• Ancillary electrical equipment (UPS losses, PDU losses, KVM switches)

Each category requires a different treatment.

IT Equipment: Nameplate vs Actual Draw

IT equipment is the dominant heat source in any server room, typically accounting for 90 to 95 percent of total sensible load. The challenge is that nameplate power ratings are worst-case figures, often set at maximum CPU utilisation, maximum memory bandwidth, and maximum storage throughput simultaneously. Real-world servers rarely sustain that load profile.

Using nameplate figures without correction leads to substantial oversizing. A 2U rack server with a 750W nameplate rating might draw 280 to 350W under typical production workloads. Across a 40-server rack, the difference between nameplate and actual draw can exceed 15 kW per rack.

Three approaches to estimating actual IT load:

• Nameplate with diversity factor: Apply a diversity factor of 0.5 to 0.7 to nameplate totals. This is a reasonable starting point for mixed-use environments but carries uncertainty.
• Measured draw: Use branch circuit monitoring, intelligent PDU data, or DCIM platform readings to capture actual watts per rack. This is the most accurate method for existing installations.
• Vendor thermal design power (TDP): Server vendors publish TDP figures that reflect sustained operational load rather than absolute maximum. TDP is more useful than nameplate for new deployments.

For design purposes, ASHRAE TC 9.9 recommends calculating IT load based on measured or TDP figures, then applying a growth margin of 10 to 20 percent depending on the planning horizon. Do not apply both a nameplate figure and a growth margin simultaneously; that compounds conservatism into a figure that no longer reflects engineering reality.

All electrical power consumed by IT equipment converts to heat at a 1:1 ratio. One kilowatt of power draw equals one kilowatt of sensible heat rejected into the space.

UPS and PDU Losses

UPS systems are not 100 percent efficient. A double-conversion UPS operating at 80 percent load typically runs at 92 to 96 percent efficiency, meaning 4 to 8 percent of input power is lost as heat within the UPS enclosure. If the UPS is located inside the cooled space, those losses add directly to the sensible load.

For a 100 kVA UPS at 0.9 power factor (90 kW input) with 94 percent efficiency, the heat rejection is approximately 5.4 kW. This figure is often omitted from preliminary calculations and then discovered during commissioning.

PDU losses are smaller, typically 1 to 2 percent of throughput, but should be included for accuracy.

Lighting

Lighting contributes a modest but measurable heat load. Fluorescent fittings in older data centres dissipate approximately 10 to 15 W/m² of floor area. LED replacements reduce this to 4 to 8 W/m². For a 200 m² server room with LED lighting at 6 W/m², the lighting load is 1.2 kW, which is small relative to IT load but worth including.

If the data centre uses motion-activated lighting that is off for most of the operational period, a demand factor of 0.1 to 0.2 is appropriate.

Building Envelope Gains

Conductive and solar heat gains through walls, roof, and glazing depend on construction type, insulation values, and orientation. In Australian climates, envelope gains can be meaningful for above-ground server rooms with external walls or roof exposure.

Conductive gain is calculated as:

Q = U × A × ΔT

Where U is the thermal transmittance (W/m²K), A is the surface area (m²), and ΔT is the temperature difference between outside and the conditioned space (K).

For a server room maintained at 24°C in Brisbane during summer, with an external wall exposed to 38°C ambient, ΔT is 14 K. A poorly insulated concrete wall (U = 2.5 W/m²K) measuring 30 m² would contribute 1.05 kW. A well-insulated wall (U = 0.4 W/m²K) reduces this to 168 W.

Solar gain through glazing requires the solar heat gain coefficient (SHGC) of the glazing and the peak solar irradiance for the orientation and latitude. Most purpose-built data centres avoid external glazing entirely, which eliminates this variable.

For internal server rooms with no external envelope exposure, envelope gains are effectively zero.

Human Occupancy

Each person present in the space dissipates approximately 90 to 120 W of sensible heat under light activity (seated or standing, performing maintenance tasks). For a server room with two technicians present during business hours, the occupancy load is roughly 200 to 240 W.

Data centres are not heavily occupied spaces. Occupancy load rarely exceeds 0.5 kW and is often excluded from the calculation without material consequence. It should still be noted in the load schedule for completeness.

Assembling the Total Sensible Load

A worked example for a mid-tier server room:

• IT equipment (measured TDP, 20 racks at 8 kW average): 160 kW
• Growth margin (15%): 24 kW
• UPS losses (200 kVA at 94% efficiency, 0.9 PF): 10.8 kW
• PDU losses (1.5% of 180 kW throughput): 2.7 kW
• Lighting (150 m² at 6 W/m²): 0.9 kW
• Envelope gains (internal room, no external exposure): 0 kW
• Occupancy (2 persons): 0.2 kW

Total sensible load: 198.6 kW

This figure becomes the basis for CRAC unit selection, but it is not the only input. The sensible heat ratio of the selected units must also be considered.

Sensible Heat Ratio and Why It Matters

Sensible heat ratio (SHR) is the proportion of a cooling unit's total capacity that handles sensible load, expressed as a decimal:

SHR = Sensible Cooling Capacity / Total Cooling Capacity

A unit with 100 kW total capacity and an SHR of 0.95 delivers 95 kW of sensible cooling and 5 kW of latent (dehumidification) capacity.

Conventional comfort air conditioning units are designed for spaces where people, cooking, and infiltration generate substantial latent loads. These units typically operate with SHR values of 0.65 to 0.75. They are optimised to remove moisture, which means a large portion of their capacity goes toward condensing water vapour rather than cooling air.

Data centres generate almost no latent load internally. IT equipment does not produce moisture. The sensible heat ratio of a server room is typically 0.95 to 1.0. If you install a comfort cooling unit with SHR of 0.70 in a server room, you are paying for 30 percent of its capacity to perform dehumidification that the space does not need. Worse, the unit will over-dehumidify the air, driving relative humidity below the lower limit of ASHRAE TC 9.9's A1 class envelope (20% RH), which creates electrostatic discharge risk.

Precision air conditioning units (CRAC units) are designed specifically for this load profile. They operate with coil temperatures above the dew point of the supply air, which means they cool without condensing. SHR values for precision cooling equipment typically range from 0.90 to 1.0 depending on configuration.

When selecting CRAC units, confirm the published SHR at your specific operating conditions: supply air temperature, return air temperature, and ambient conditions. Manufacturers publish SHR curves across operating points; the catalogue SHR at standard conditions may differ from SHR at your actual design conditions.

For the 198.6 kW example above, if you select units with SHR of 0.92 and total rated capacity of 220 kW, the effective sensible capacity is 202.4 kW, which provides adequate margin. If you select units with SHR of 0.75 (comfort units), the effective sensible capacity from 220 kW total is only 165 kW, which is insufficient despite the headline capacity figure appearing adequate.

Applying N+1 or 2N Redundancy

Sensible load calculation gives you the minimum cooling capacity required. Redundancy requirements then determine how many units you need and at what individual capacity.

For N+1 redundancy with a 198.6 kW load and units rated at 70 kW sensible capacity each, you need a minimum of three units to meet load (210 kW), plus one standby: four units total. Each unit must be capable of carrying its share of the load with one unit offline.

This is a separate calculation from the load estimate itself, but it depends on the load estimate being accurate. An inflated load figure drives unnecessary capital expenditure on additional cooling units and the infrastructure to support them.

Putting the Calculation to Work

Sensible heat load calculation is not a one-time exercise. As IT equipment changes, rack densities increase, or the building envelope is modified, the load profile shifts. Facilities teams that maintain a live load schedule, updated against measured PDU data, are better positioned to make informed decisions about cooling capacity additions or reconfigurations.

Thermal modelling tools, including computational fluid dynamics (CFD) analysis, can validate calculated loads against predicted airflow behaviour. This is particularly useful in high-density environments where localised hot spots may develop even when overall cooling capacity is adequate.

For server rooms and data centres across Brisbane, Sydney, and Melbourne, CRAC Services Australia provides load assessments, thermal modelling, and CRAC unit selection as part of cooling design engagements. If you are working through a sizing exercise or reviewing an existing installation, visit [https://crac.services](https://crac.services) to discuss your requirements.