1-833-888-0091contact@hvacengineering.ca

Chilled water plant design for data centers

Explains how a data center chilled water plant differs from a comfort cooling plant, and the design choices that set its efficiency and reliability.

CRAH Rack Cold aisle 18-27 °C Hot aisle Return plenum Liquid rack CDU
Data center cooling and HVAC design

Most large data centers reject heat through a chilled water plant. Chillers, pumps, cooling towers or dry coolers, and heat exchangers produce water that serves CRAH units, air handlers, in-row coolers, rear doors and coolant distribution units. Compared with a comfort cooling plant, a data center plant runs continuously, carries a nearly constant load, operates at higher water temperatures and must keep running through equipment failures and maintenance.

Those differences change many design decisions. This guide covers the main ones.

How a data center plant differs from a comfort plant

FactorComfort cooling plantData center plant
Load profileVaries with weather and occupancyNear constant, driven by IT load
Operating hoursSeasonal, occupied hours24/7/365
Chilled water supplyOften 6 to 7 °C (43 to 45 °F)Often 12 to 20 °C (54 to 68 °F) or warmer
Latent loadSignificantSmall; mostly sensible
Winter coolingLowFull load, so economizer potential is large
RedundancyN or modestN+1, 2N or concurrently maintainable
Failure consequenceDiscomfortIT outage

Water temperatures

Because data center loads are almost entirely sensible, chilled water does not need to be cold enough to dehumidify. Supply temperatures in the mid-teens °C are common with CRAH units, and warmer still with close-coupled cooling or liquid cooling. Higher temperatures bring three benefits:

  1. Chillers operate at lower lift and higher efficiency.
  2. Waterside economizers carry more hours of the year.
  3. Coils run dry, avoiding condensate and unintended dehumidification.

Design delta T matters as well. A wider delta T, often 8 to 12 K in modern designs, reduces flow and pump energy and lets pipes be smaller. Achieving it requires coils and controls designed for it, and attention to low delta T syndrome.

Chiller selection

Water-cooled centrifugal chillers, including magnetic-bearing variable-speed machines, are common in large plants because of their efficiency at the low lift data center plants enjoy. Air-cooled screw or scroll chillers with integral free cooling coils are common in smaller or water-constrained sites. Key selection points:

  • Part-load and low-lift efficiency at the actual operating temperatures, not just at AHRI rating conditions
  • Fast restart after a power interruption, measured in seconds to minutes depending on the machine
  • Refrigerant choice in light of federal and provincial regulations and CSA B52 machinery room requirements. See refrigerant regulations and low-GWP options.
  • Turndown, so a chiller can operate stably at light loads during early phases of build-out

Economizer integration

In Canada, the waterside economizer is a central part of the plant, not an add-on. A plate heat exchanger in series ahead of the chillers precools return water whenever the condenser or glycol loop is colder than chilled water return. Chillers then trim the remainder. In full economizer mode, chillers turn off entirely.

Dry coolers with glycol avoid tower freezing and water use; open towers with winter operation provisions offer closer approaches. See free cooling and economizers for data centers in Canada.

Pumping arrangement

Variable primary flow is common in new data center plants because it removes a pumping stage and suits a stable load. It requires flow measurement, minimum flow bypass and careful control during chiller staging. Primary-secondary arrangements remain common where chillers have strict flow requirements or the plant serves multiple loads with different needs. See primary-secondary vs variable primary flow.

Redundancy and distribution

Plant redundancy is set by the owner's availability requirements and often by an Uptime Institute tier objective. Common approaches include:

  • N+1 chillers, pumps and towers on common headers
  • Dual distribution paths (A and B loops) to every load, so piping maintenance does not interrupt cooling
  • Ring mains with sectional isolation valves
  • 2N plants with fully independent systems in the highest-availability designs

Piping must allow any component to be isolated for maintenance without affecting cooling. That means valves on both sides of every major component, and valves at strategic points on headers. See data center cooling redundancy.

Thermal storage and ride-through

When utility power fails, generators start within seconds, but chillers may take several minutes to restart and reach capacity. During that gap, IT equipment on UPS keeps generating heat. Chilled water thermal storage tanks, sized for the restart time plus margin, keep cooling running. Pumps and CRAH fans are usually on UPS so that stored chilled water keeps circulating.

Water treatment and cleanliness

Closed chilled water loops need treatment for corrosion and biological control, plus side-stream filtration. If the plant serves direct-to-chip CDUs, the facility loop does not contact cold plates directly, but clean water still protects heat exchangers and control valves. Open condenser loops need a full treatment and Legionella management program.

Controls and sequencing

Data center plant controls should:

  1. Stage chillers, pumps and towers to minimize total plant power, not just chiller power
  2. Manage transitions between economizer, partial economizer and mechanical modes smoothly
  3. Reset chilled water supply temperature upward when loads allow
  4. Handle failures automatically, starting standby equipment without operator intervention
  5. Restart the plant automatically after power loss in a defined sequence
  6. Provide clear alarms and trends to the BAS and DCIM

Sequencing logic should be documented clearly in the sequence of operation and verified during commissioning.

Phased build-out

Many facilities open with a fraction of the design IT load. Plants should be designed for staged installation, with header connections, pads and space for future chillers. Selecting equipment that runs efficiently at low initial load avoids years of poor performance.

Frequently asked questions

Air-cooled or water-cooled chillers?

Water-cooled plants are usually more efficient and suit large sites, but they use water and need towers. Air-cooled chillers with free cooling coils are simpler, use no water and are common in smaller or water-restricted sites.

How warm can chilled water be?

That depends on the terminal units. CRAH coils and rear doors are selected for a given entering water temperature; raising it reduces their capacity unless they were sized for it. Plan warm water temperatures from the start.

Key takeaways

  • Data center plants run continuously at near-constant, mostly sensible loads.
  • Warmer chilled water and wider delta T improve efficiency and economizer hours.
  • Waterside economizers are central to Canadian plant design.
  • Redundancy, isolation valves and dual paths support maintenance without downtime.
  • Thermal storage and fast-restart chillers bridge the gap after power interruptions.

Talk to an engineer

Have a project or a problem this guide touches on? Tell us about the building and we’ll be in touch within one hour.