In brief
A coolant distribution unit (CDU) transfers heat between the technology cooling system (TCS) and facility water system (FWS). Size it against duty, supply and return temperatures, fluid properties, pressure drops and the capacity required during faults and maintenance.
The Coolant Distribution Unit is the single most important piece of equipment in a liquid-cooled data center. It separates the cooling loop that serves the servers from the facility water system, and it holds that server loop’s temperature, flow and pressure at their setpoints regardless of what happens on the facility side.
Without that separation, the narrow flow channels inside the servers would be exposed to facility loop pressure, contamination and temperature swings. A CDU is therefore a protection device, a control device and a pumping station at the same time. The wider picture is covered in our complete guide to data center cooling.
What a buyer should provide
For a defined installation or commissioning scope, share the CDU make and model, unit count, P&ID, FWS/TCS connections, design duty, temperature limits, schedule and test and warranty responsibilities.
How a CDU works
The CDU sits between two loops. The Facility Water System (FWS) brings cooling water from heat rejection, whether dry coolers or chillers. The Technology Cooling System (TCS) carries coolant into the racks and on to the cold plates on the processors.
The unit transfers heat from one loop to the other through a heat exchanger while handling four jobs at once:
Temperature control. The TCS supply temperature is held within the range the hardware manufacturer specifies, by modulating facility water flow through the heat exchanger. It must also stay above the hall’s dew point so that moisture does not condense on cold pipework and connectors.
Flow. Pumps circulate coolant through the server loop and hold flow or differential pressure at setpoint. Variable speed drives give the pumps the ability to follow load, which in AI compute can change fast and by a lot.
Pressure. Specify TCS and FWS pressures within OEM limits and for credible leak scenarios. An internal heat-exchanger leak flows from higher pressure to lower pressure. If FWS pressure is higher, FWS fluid could enter TCS. Plan detection, isolation and fluid compatibility.
Fluid condition and volume. Filtration removes the particles that would otherwise block cold plate channels. The expansion and fill system handles thermal expansion and air removal.
On top of these, a CDU carries its own instrumentation, alarms and leak detection, plus an interface to the building management or DCIM system.
CDU types
Floor-mounted liquid-to-liquid (L2L): connects one or more rack groups to FWS. Verify capacity, maintenance access and resilience.
Row- or rack-mounted L2L: close to the load. Verify space and plant-side connections.
Liquid-to-air (L2A): rejects heat into room air. The hall’s air-cooling system must carry that load.
The three numbers that decide the sizing
Load and temperature difference. Transferred power follows the product of flow and temperature difference. A one megawatt load across a ten degree difference means roughly 86 cubic metres per hour of water. Halve the temperature difference and the flow doubles, while pipe size, pressure loss and pumping power require separate hydraulic calculations. The design temperature difference is therefore the single most consequential choice in the sizing.
Heat exchanger approach. The server loop supply temperature cannot be lower than the facility loop supply temperature. Between them sits the heat exchanger approach, typically a few degrees. If the hardware requires 32 °C supply and the approach is three degrees, the facility loop has to deliver water at no more than 29 °C. That one number determines how much of the year the site runs on free cooling and how valuable the recovered heat is. An unnecessarily tight approach requirement can sink the economics of heat recovery before a single heat pump has been specified.
Fluid properties. A glycol mix lowers specific heat capacity and raises viscosity, so the same duty needs more flow and more pump power than plain water would. Glycol content is a sizing parameter, not a freeze protection decision made afterwards.
Alongside these, the sizing has to establish the total pressure drop of the server loop, meaning manifolds, hoses, quick disconnects and cold plates together, and the redundancy level. Pump redundancy inside a unit does not substitute for system level redundancy when one CDU serves several rack rows.
Installation and interfaces
A CDU is heavy, position-critical and connected in several ways at once. Installation has to resolve floor loading and anchoring, service clearance around the unit, pipe connections and their quality, electrical supply usually from two separate feeds, and the interface to automation and leak detection.
The most critical single item is cleanliness. The server loop is flushed and verified clean before it is connected to the cold plates. Weld scale, swarf or gasket debris left from installation otherwise ends up directly in a processor cooling channel, where it can cause a blockage or damage that may require component service or replacement. The same applies to quick disconnects, where handling and protection during installation translate directly into later leak risk.
Material and fluid chemistry questions are covered in the article liquid cooling water chemistry, filtration and material compatibility, and loop architecture and scope splits in TCS and FWS: cooling loop architecture and scope splits.
Commissioning
Commissioning is where sizing and installation errors become visible. Bringing a liquid-cooled loop into service typically includes:
- Pressure and leak testing before connecting IT equipment.
- Flushing, filling with the specified fluid and air removal.
- Verifying fluid cleanliness, filtration and pump and drive operation.
- Measuring and balancing flow loop by loop.
- Testing temperature response, pump and CDU faults, power transfer, leak detection and alarms.
Flow verification and balancing are in practice done with a portable clamp-on ultrasonic meter, because it measures from outside the pipe without opening the loop and one instrument covers every measurement point. The differences between measurement technologies are covered in the complete guide to industrial flow meters.
Commissioning is also the moment to record the baseline that every later measurement is compared against: flows per loop, differential pressures, temperature differences and pump vibration levels. Without a baseline, later measurements give you a value but not a trend.
Common faults
Blocked filters and flow channels. Shows up as rising filter differential pressure and falling flow in individual loops. The cause is almost always a shortfall in installation or flushing, or particles generated by corrosion.
Air in the loop. Causes uneven flow, noise, poor heat transfer and, at worst, pump cavitation. It usually traces back to incomplete air removal during filling or a leaking point under suction pressure.
Condensation. If the supply temperature drops below the hall dew point, moisture condenses on pipework and connectors. The problem is a control problem rather than a mechanical one, and is fixed with setpoints and dew point monitoring.
Control hunting. An oversized control valve or a badly tuned loop makes the temperature oscillate as load changes. The phenomenon is familiar from process industry and so is the remedy: correct valve sizing and tuning. The subject is covered in the article on control valves in industry.
Pump wear. Bearing and seal damage develops gradually and is detectable by vibration measurement weeks before failure.
Fluid degradation. Inhibitor depletion, microbial growth and conductivity drift progress slowly and are noticed only if the fluid is analysed on a schedule.
Facility side temperature drift. If FWS temperature rises above design, the CDU can no longer reach its setpoint even though the unit itself is healthy. The fault is in heat rejection, not in the CDU.
Maintenance and condition monitoring
CDU maintenance rests on four things: monitoring and changing filtration, analysing the fluid regularly, monitoring pump and valve condition, and calibrating the instrumentation. Leak detection should be function-tested at intervals, because it is a system whose failure is otherwise discovered only when there is a leak.
For the rotating machinery the practice is the same as in any other critical process. Continuous or wireless vibration monitoring on the pumps, and machinery protection on the most critical machines, buy time to react before capacity is lost. The available solutions are on our safety automation and condition monitoring page.
Finkova’s scope and next step
Finkova installs and commissions data center cooling in the Nordics, with instrumentation and lifecycle support. The work scope, OEM requirements and acceptance criteria are agreed for each project.
Contact Finkova with the project inputs above to discuss the relevant installation, testing and handover interfaces. See the Finkova delivery scope for the practical work package.
Frequently asked questions
What is a CDU?
A CDU, Coolant Distribution Unit, separates the cooling loop serving the servers from the facility water loop. It holds the server loop’s temperature, flow and pressure at setpoint and protects the narrow cold plate channels from facility loop pressure and contamination. It is the single most important piece of equipment in a liquid-cooled data center.
What is inside a CDU?
A CDU contains a heat exchanger, pumps, filtration, an expansion and fill system, and controls. It also carries instrumentation, alarms, leak detection and an interface to the building management or DCIM system. The pumps are typically on variable speed drives so they can follow rapidly changing load.
How should the TCS/FWS pressure differential be set?
There is no universal pressure order. A heat-exchanger leak follows the differential from higher pressure to lower pressure. Set TCS and FWS pressure limits, detection and isolation to the OEM requirements and the risk of fluid mixing.
What is the difference between L2L and L2A units?
A liquid-to-liquid unit (L2L) transfers heat into the facility water loop and suits purpose-built liquid-cooled halls. A liquid-to-air unit (L2A) transfers heat straight into hall air with no facility water connection. It is a quick way to start, but it does not remove heat from the hall, only hands it to the air cooling.
How is a CDU sized?
Three numbers decide the sizing: load and temperature difference, heat exchanger approach, and fluid properties. The total server loop pressure drop and the redundancy level are also defined. Temperature difference matters most, because halving it doubles flow for the same duty. Pumping power also depends on pressure loss and efficiency.
What does heat exchanger approach mean?
Approach is the temperature difference between the two sides of the heat exchanger, typically a few degrees. The server loop supply temperature cannot be lower than the facility loop supply. If the hardware requires 32 °C and the approach is three degrees, the facility loop must deliver water at no more than 29 °C.
How much flow does one megawatt of cooling need?
A one megawatt load across a ten degree temperature difference needs roughly 86 cubic metres per hour of water. Halve the temperature difference and flow doubles for the same duty. Pipe size and pump power need separate hydraulic calculations. A glycol mix needs more flow than plain water for the same duty.
What are the most common CDU faults?
Blocked filters and flow channels, air in the loop, condensation from too low a supply temperature, control hunting, pump wear and fluid degradation. A rise in facility loop temperature can also stop a CDU reaching its setpoint even though the unit itself is healthy.
Why is flushing important in CDU installation?
Weld scale, swarf or gasket debris from installation otherwise ends up directly in a cold plate channel, where it does damage no later flush will undo. The server loop is therefore flushed and verified clean before it is connected to the cold plates. This is the single most critical installation step.
Is pump redundancy inside a CDU enough?
Not necessarily. Internal pump redundancy protects against a pump failure, but not against a failure of the whole unit. If one CDU serves several rack rows, its failure stops all of them. Redundancy therefore has to be designed at system level as well.
