Data center cooling pumps serving chilled water and liquid cooling loops
A data center is not cooled by a single pump. It is cooled by several pump groups working in series, each one moving heat across a different boundary in the thermal chain: from the server, to the room, to the chiller, and finally to the outdoors. Understanding data center cooling pumps means understanding those boundaries first, then matching pump duty to the loop it serves.
Not every facility uses every group. A small edge site with packaged DX units may have almost no process pumps at all, while a hyperscale campus with liquid-cooled GPU halls may run four or five distinct pumping systems in parallel. The sections below map where pumps fit, how their roles differ, and what changes when racks move from air to liquid cooling.
الوجبات الرئيسية
- A data center cooling system is a chain of loops, and each loop usually needs its own pump group sized for that loop’s flow, head, and temperature.
- The most common groups are chilled water pumps, condenser water pumps, coolant distribution unit (CDU) and secondary-loop pumps, and heat-rejection or economizer pumps.
- Liquid cooling at the rack (direct-to-chip or immersion) does not replace facility pumps; it adds a low-temperature-rise secondary loop that still rejects heat to a facility water system (FWS).
- Pump selection hinges on flow rate, total dynamic head, NPSH available, water quality, and how the loop is controlled, not just nameplate capacity.
- Reliability comes from redundancy (N+1 or 2N), variable frequency drive (VFD) control, clean water chemistry, and pumps that can be serviced without dropping the load.
Where Pumps Fit in the Data Center Cooling Chain
Heat in a data center moves outward through a series of loops. Servers reject heat into the room air or into a cold plate. That heat is then picked up by a CRAH coil or a CDU heat exchanger and handed to a chilled water loop.
The chilled water loop carries it to a chiller evaporator. The chiller lifts the heat to a condenser water loop, which delivers it to a cooling tower, dry cooler, or adiabatic unit that releases it to ambient air. In economizer mode, parts of this chain may be bypassed when outdoor conditions allow.
Each handoff between loops is a closed or open hydronic circuit, and each circuit needs a pump set sized for its own duty. Grundfos’s introduction to data center cooling walks through this layered view, and ASHRAE’s data center HVAC handbook chapter treats each loop as a distinct design problem. The role-based table below summarizes the typical groupings.
Pump group | Loop served | Open or closed | Typical duty | Heat source it sees |
|---|---|---|---|---|
Primary chilled water | Chiller evaporator loop | Closed | Constant or near-constant flow at low head | Returns from CRAHs / CDUs |
Secondary chilled water | Distribution to CRAHs and CDUs | Closed | Variable flow, moderate-to-high head | Room and rack cooling units |
Condenser water | Chiller condenser to cooling tower | Open (towers) | High flow, moderate head with static lift | Chiller condenser |
CDU / TCS secondary | Rack-level technology cooling system | Closed | Low head, tight ΔT, treated water | Cold plates or immersion tanks |
Heat-rejection / economizer | Dry coolers, adiabatic units, free-cooling coils | Closed or open | Variable, weather-dependent | Outdoor heat exchanger |
Main Types of Data Center Cooling Pumps
Chilled Water Pumps
Chilled water pumps move cold supply water from the chiller plant to the cooling units in the white space and return warmer water back. Most large facilities split this into two stages.
Primary chilled water pumps serve the chiller evaporators. They are typically constant-speed or have a narrow turndown window because chillers need predictable evaporator flow to stay efficient and avoid freeze trips. Head requirements are modest, since the loop is short and the chiller barrel is the dominant friction element.
Secondary chilled water pumps push water through the distribution headers to CRAHs, CRACs, and CDU primary sides. This loop sees variable load: a row that is half-populated draws less flow than one running AI training jobs at full power. Secondary pumps are almost always equipped with VFDs and controlled on differential pressure at the hydraulically remote unit, so flow tracks the actual cooling demand.
Total dynamic head here is higher than the primary loop because of long pipe runs, control valves, and coil pressure drops.
A primary-secondary (decoupled) arrangement isolates the two loops with a common pipe so that variable secondary flow does not destabilize the chillers. Some newer designs use variable primary flow with a minimum-flow bypass, which removes one set of pumps but demands tighter controls.
Condenser Water Pumps
Condenser water pumps move warm water from the chiller condensers up to the cooling towers and back. This is usually an open system: water is exposed to atmosphere in the tower basin, so the pumps have to overcome both friction loss and the static lift from the basin water level to the spray nozzles or distribution deck.
Because the loop is open, two design points matter more than in closed systems. First, NPSH available has to comfortably exceed NPSH required, especially when towers sit only slightly above the pumps. Second, water quality is harsher: airborne dust, biological growth, and dissolved solids concentrate as water evaporates, so pumps need corrosion-resistant materials and the loop needs filtration and chemical treatment.
ASHRAE’s condenser water systems chapter covers sizing, materials, and water treatment for these systems in detail.
Condenser water pumps are typically end-suction or split-case centrifugal units sized for high flow at moderate head. They are often staged: one pump per operating chiller, plus at least one standby.
CDU and Secondary-Loop Pumps
When racks use direct-to-chip liquid cooling or immersion cooling, the heat does not leave the rack as warm air; it leaves as warm water or dielectric fluid. A coolant distribution unit sits between the facility water system (FWS) and the technology cooling system (TCS). Inside the CDU, a brazed-plate or shell-and-tube heat exchanger transfers heat from the TCS to the FWS, and a dedicated pump set circulates the TCS side.
These pumps look different from facility pumps. Flow per kilowatt is high but head is low, because TCS piping is short and the cold plates have engineered pressure drops. ΔT is small (often 5–10 °C) so any drop in flow shows up immediately as a rise in chip temperature. Water chemistry is controlled tightly: deionized or treated water, sometimes with propylene glycol, filtered to single-digit microns to protect microchannel cold plates.
Redundant pumps inside the CDU are standard, usually configured so that one pump can fail or be serviced while the other carries the full load. Variable speed control adjusts TCS flow to track IT load and keep ΔT in the design band.
Heat-Rejection and Economizer Pumps
Not all heat leaves through a chiller. Many facilities use waterside economizers, dry coolers, or adiabatic coolers to reject heat directly to ambient when wet-bulb or dry-bulb temperatures allow. These paths have their own pump groups.
A waterside economizer typically uses a plate-and-frame heat exchanger in parallel with the chiller. When outdoor conditions are cold enough, condenser water (or a separate glycol loop) cools the chilled water loop without running compressors. The condenser water pumps may continue to serve this duty, or a dedicated economizer pump may be installed.
Dry-cooler and adiabatic-cooler loops are closed glycol circuits with their own pumps, sized for the worst-case ambient day. Because these pumps may run year-round at varying loads, VFD control and staged operation are standard. Grundfos catalogs a range of pumps for data center applications covering both chilled water and heat rejection duties.
How Pump Requirements Change With Liquid Cooling
Air-cooled halls and liquid-cooled halls put very different demands on pumps. In an air-cooled facility, CRAHs run on chilled water at roughly 7–12 °C supply, and pump selection is dominated by the long distribution runs to many small coils. ΔT across the coil is typically 5–7 °C, so flow per kilowatt is high.
Liquid cooling shifts the picture in several ways:
- Higher supply temperatures. Direct-to-chip and immersion systems often accept FWS supply temperatures of 25–40 °C. That means chillers may run partially loaded or be bypassed entirely in favor of dry coolers, changing which pump groups carry the duty.
- Two stacked loops instead of one. The TCS loop inside the rack and the FWS loop between the CDU and the heat-rejection plant each need their own pumps with very different specs.
- Tighter flow control. Cold plates have narrow operating windows. A pump that hunts or oscillates on a poorly tuned VFD will show up as chip-temperature noise long before it shows up in the building automation system.
- Cleaner water, stricter materials. TCS-side pumps see treated water with low conductivity. Stainless steel, engineered plastics, and EPDM seals are common; bronze or cast iron may not be acceptable.
A facility that adds a liquid-cooled hall to an existing air-cooled plant rarely replaces its chilled water pumps. It adds CDU pumps and may add a dedicated heat-rejection loop, while the original chilled water system continues to serve the air-cooled portion.
Pump Selection Factors That Matter
Sizing a data center cooling pump is more than picking a curve that passes through the design point. The factors that drive long-term performance include:
- معدل التدفق at design load, plus the realistic turndown range as IT load varies day to day.
- Total dynamic head, broken down into static lift (for open loops), friction loss in pipe and fittings, and pressure drop across coils, heat exchangers, and control valves.
- NPSH available versus required, particularly for condenser water pumps drawing from tower basins and for any pump handling warm water near saturation.
- Efficiency at part load, not just at the duty point. Most pumps spend most of their hours at 40–70 % of design flow.
- VFD compatibility and control strategy, including whether the pump is controlled on differential pressure, ΔT, or a hybrid scheme.
- Materials and water chemistry matched to whether the loop is open or closed, treated or untreated, glycol or pure water.
- Footprint and maintainability, since pumps that cannot be isolated and serviced without dropping the load will eventually force a planned outage.
Open versus closed hydronic systems deserve a specific call-out. Open loops (condenser water, some heat-rejection circuits) contend with oxygen, biological growth, and static lift. Closed loops (chilled water, TCS) are easier on materials but punish any air ingress with cavitation and corrosion.
Reliability, Controls, and Maintenance
Data center pump systems are designed around the assumption that any single component can fail without taking the load down. N+1 redundancy is the floor for most enterprise facilities; 2N or distributed redundancy is common at the hyperscale tier. Pumps are arranged in parallel with isolation valves, check valves, and headers that allow one unit to be pulled for service while the standby carries the duty.
Controls tie the pumps to the rest of the plant. Secondary chilled water pumps modulate on remote differential pressure. CDU pumps modulate on TCS ΔT or cold-plate inlet temperature.
Condenser water pumps stage with chillers and tower cells. A well-tuned sequence keeps pumps off their end stops and away from low-flow regions where bearings and seals wear quickly.
Maintenance habits that pay back over a facility’s life include keeping water chemistry inside the treatment program’s specification, replacing strainer baskets and filters on schedule, monitoring vibration and bearing temperature for early warning, and rotating lead-lag duty among parallel pumps so that wear is shared evenly.
الأسئلة الشائعة
Do all data centers need chilled water pumps?
No. Small or edge facilities using packaged DX or split air conditioners may have no process pumps beyond a small condensate pump. Chilled water pumps appear once a central plant is justified.
How are primary and secondary chilled water pumps different?
Primary pumps serve the chiller evaporators at near-constant flow and modest head. Secondary pumps serve the distribution to CRAHs and CDUs at variable flow and higher head, almost always on VFDs.
Where do CDU pumps sit in the chain?
Inside the coolant distribution unit, on the technology cooling system (TCS) side. They circulate treated water or coolant between the CDU heat exchanger and the cold plates or immersion tanks, while the facility water system (FWS) on the other side of the heat exchanger is served by the chilled water or heat-rejection pumps.
Is liquid cooling replacing chilled water pumps?
No. Liquid cooling adds a new secondary loop and its own pumps; it does not remove the need to reject heat from the building. Many liquid-cooled designs still use a facility water loop with conventional pumps, though warmer supply temperatures may shift duty from chillers to dry coolers.
What is the most common reason a data center pump fails early?
Water-side issues: poor chemistry, fouled strainers, or air entrainment that drives cavitation and seal damage. Mechanical failures usually trace back to operating conditions outside the pump’s intended range.
الخاتمة
Data center cooling pumps are best understood as a set of role-specific machines, not a single product category. Chilled water pumps move heat from the white space to the chillers. Condenser water pumps lift it from the chillers to the towers.
CDU and TCS pumps handle the short, low-head, high-cleanliness loops inside liquid-cooled racks. Heat-rejection and economizer pumps connect the plant to outdoor air whenever conditions allow. Selecting each group on its own loop’s flow, head, water quality, and control strategy, and backing the selection with sensible redundancy and maintenance practices, is what keeps the thermal chain intact from chip to sky.