The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Liquid cooling does not automatically keep running when a data center loses utility power. Coolant circulation continues only if the pumps, controls, and heat-rejection equipment have UPS, generator, or redundant power available. If circulation or heat rejection stops while servers remain powered, equipment temperatures can rise very quickly. Depending on the design, the site may maintain cooling, switch to a redundant pump or coolant distribution unit (CDU), reduce IT load, or shut servers down in a controlled sequence.
There is no universal outage response for every liquid-cooled data center. The result depends on the entire cooling chain, not merely whether a CDU has a UPS connection.
Why a power outage affects more than the server pump
A typical liquid-cooled installation has several linked parts. A CDU circulates a technology cooling loop through cold plates or another server heat-transfer interface. Inside the CDU, a heat exchanger transfers that heat to a facility loop or to another heat-rejection path. Pumps, valves, sensors, control electronics and communications keep the system operating within its design limits.
The facility side may include water pumps, chillers, dry coolers, cooling towers, air handlers or other heat-rejection equipment. A server can therefore remain connected to a powered CDU while losing the facility equipment that removes heat from the building. Conversely, facility cooling may remain available while a CDU pump or controller is unpowered.
#1 Best Overall
ASHRAE guidance describes CDU implementations as variable rather than standardized. Eaton’s descriptions likewise identify the secondary loop, cold plates, heat exchanger and monitoring of temperature, pressure and flow as parts of a coordinated system. A UPS label on one cabinet does not prove that every required component will operate during an outage.
Four things that can happen when utility power is lost
1. Cooling continues for the event
A UPS or generator-backed supply can keep CDU controls and pumps operating while the standby electrical system starts or while the outage lasts. Automatic transfer switches (ATS) can move a load to an alternate source without waiting for a manual intervention. The duration depends on battery capacity, generator fuel and the power demand of the complete cooling chain.
Some commercial designs illustrate this approach. Trane documents a CDU-1MW design with an ATS, UPS-backed controller continuity, dual concurrently maintainable pumps and fault handling intended to avoid an immediate shutdown. Those are characteristics of that product design, not a guarantee for other CDUs or for the facility equipment connected to it.
Recommended Free Tools
2. A redundant pump or CDU takes over
Liquid-cooling systems can be built with standby pumps, dual CDUs or separate power feeds. Controls may detect a pump fault, loss of flow or abnormal pressure and start the alternate component automatically. Vertiv describes automated failover for redundant components and units, while ASHRAE recommends fault-tolerant CDU design and backed-up CDU power.
Automatic failover is useful only if the standby path is actually available. The alternate pump may share a failed switchboard, the second CDU may depend on the same facility loop, or a transfer sequence may require a control network that has also lost power. Operators should verify the electrical and hydraulic independence of each supposedly redundant path.
3. Flow or heat rejection is interrupted
If a required pump, controller or heat-rejection device loses power without a working alternate, active cooling can be reduced or stop. Servers that remain powered continue producing heat. ASHRAE warns that datacom equipment temperature can rise very quickly if a CDU fails.
No universal time-to-overheat value can be applied to all systems. The rate depends on server load, coolant volume, cold-plate design, room conditions, allowable inlet temperature, control setpoints and whether residual or passive heat transfer remains. A number measured on one installation should not be used as an emergency limit for another.
4. IT load is reduced or shut down
If cooling continuity cannot be restored, operators may need to shed compute load or shut servers down before thermal limits are reached. ASHRAE says CDU power should be backed by a UPS so equipment can either continue running or shut down gracefully during an outage. The trigger, order and timing must come from the site’s controls and operating procedures; they should not be improvised during an incident.
Rank #3
Liquid-to-liquid and liquid-to-air systems respond differently
Liquid-to-liquid cooling
In a liquid-to-liquid arrangement, the CDU transfers heat from the technology loop to facility water. The technology pump, CDU controls and facility-side pumps must all remain available. A loss of chilled-water production, cooling-tower operation or facility-water circulation can stop heat rejection even when the server-side pump is still turning.
For this architecture, the outage plan must identify every electrical load between the server cold plate and the final heat sink. That normally includes CDU pumps and controls, facility pumps, chillers or other heat-rejection equipment, and the controls that coordinate them.
Liquid-to-air cooling
A liquid-to-air CDU rejects heat through the data-center air-conditioning system rather than a separate facility-water loop. The CDU may continue circulating coolant while relying on air handlers or another air-side system to remove the heat. If those air handlers lose power, the liquid loop can remain full and moving but still fail to reject heat effectively.
Free tools Windows power users keep installed
One-click scans. No signup required.
Because the two architectures have different dependencies, a power study that covers a liquid-to-liquid installation cannot simply be copied for a liquid-to-air room. CDU placement, loop arrangement, valve logic, controls and backup feeds all affect the sequence.
Rank #4
What backup power must cover
Resilience requires a system-wide review. The following table shows the questions that matter during an outage:
| Cooling element | What must be confirmed | Typical failure consequence |
|---|---|---|
| CDU pumps | UPS or generator feed, runtime, pump redundancy and automatic start | Loss or reduction of technology-loop flow |
| CDU controller and sensors | Backed-up power, control-network availability and alarm reporting | No automatic transfer, monitoring or safe shutdown command |
| Facility-side pumps | Separate electrical path and compatibility with the standby source | Heat exchanger has no usable facility flow |
| Chillers, towers, dry coolers or air handlers | Whether the required heat-rejection equipment remains online | Coolant circulates but heat accumulates |
| ATS and switchgear | Transfer sequence, interlocks, test results and load capacity | Delayed or failed movement to standby power |
| Telemetry and alarms | Power-independent communications and operator notification | Operators discover thermal risk too late |
Backup runtime must be calculated for the actual load. A small UPS that keeps a controller alive for several minutes may not run a large pump, and a generator that supports IT racks may not have capacity for chillers or cooling towers. The design objective may be continuous operation, a controlled transfer, or enough time for an orderly IT shutdown.
How operators should prepare before an outage
- Map the complete cooling chain. Document the path from each server cold plate through the technology loop, CDU, heat exchanger, facility loop and final heat sink. Mark every pump, valve, controller and electrical panel.
- Record power dependencies. For each item, identify the UPS, generator bus or utility feed, expected runtime and breaker. Confirm whether an ATS transfers it automatically and whether transfer affects pump direction, control state or valve position.
- Test redundancy under realistic conditions. Verify that a standby pump or CDU starts, reaches the required flow and reports healthy status. Test failure of the primary component, not only a software command, and confirm that both units are not dependent on one unprotected panel.
- Define thermal and operational thresholds. Establish the flow, pressure, supply-temperature and return-temperature limits that trigger an alarm, load reduction or shutdown. Tie each threshold to a named action and responsible operator.
- Protect telemetry. Ensure that temperature, pressure, flow, pump status and power alarms reach the operations team when normal network equipment is unavailable. Remote visibility should remain available on the same standby-power plan as the cooling controls.
- Exercise the shutdown sequence. Confirm which workloads are stopped first, how orchestration drains them, and how servers are powered down without leaving pumps or valves in an unsafe state. Record the manufacturer’s instructions for the CDU and connected equipment.
What to do during the outage
When cooling remains healthy
Confirm that the standby source has accepted the load, that CDU pumps are running, and that flow and temperatures remain within the site’s limits. Watch for a delayed loss of facility heat rejection: a system can appear normal immediately after transfer and warm later if a chiller, tower or air handler did not start.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →When a redundant path is transferring
Check the alarm that initiated the transfer, the identity of the active pump or CDU, and the achieved flow. Do not return the failed component to service merely because utility power has returned; investigate the original fault and follow the equipment procedure.
Best Value
- Data Center Coolant
- 25% Inhibited Propylene Glycol
- JeffCool ISF 25
- High thermal conductivity
When flow or heat rejection is falling
Reduce nonessential IT load according to the documented sequence. If temperatures continue moving toward the site’s emergency threshold, begin the controlled server shutdown rather than waiting for an automatic thermal trip. Keep records of alarms, transfer times and temperatures for post-incident analysis.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Restoring service after power returns
- Verify that utility power is stable and that the standby system is not still carrying the load unexpectedly.
- Inspect CDU, pump, valve and facility-equipment alarms before restarting anything.
- Restore the cooling chain in the manufacturer-specified order, normally establishing facility heat rejection before adding full server load.
- Confirm stable flow, pressure and supply and return temperatures at the CDU and at representative server interfaces.
- Return compute load gradually, watching for a delayed temperature increase or a second pump starting unexpectedly.
- Close the incident only after telemetry, alarms, redundancy and automatic transfer functions have been tested or formally accepted as unavailable.
How to compare resilience designs
When evaluating a new installation or reviewing an existing one, compare the following dimensions rather than asking whether a system is simply “UPS-backed.”
| Comparison axis | Questions to ask |
|---|---|
| Backup scope | Are only controls backed up, or do pumps and all required heat-rejection devices receive standby power? |
| Runtime | How long can batteries and generators support the cooling load at the planned outage condition? |
| Redundancy | Are pumps, CDUs, power feeds and facility equipment independently redundant? |
| Failover | Does transfer occur automatically, and what alarm confirms that the alternate path is carrying the load? |
| Monitoring | Are flow, pressure, temperature, power and leak alarms visible when normal communications are down? |
| IT response | Can orchestration reduce load or shut servers down gracefully before thermal limits are reached? |
| Recovery | Is there a documented restart and load-restoration sequence for this exact CDU and facility design? |
Trane and Eaton document different examples of redundant pumps, backup controls and transfer features. Those examples demonstrate available design choices, not a ranking of all systems or proof that one architecture will perform better in every data center.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Common assumptions that lead to mistakes
- “The servers have UPS power, so cooling is covered.” IT UPS capacity says nothing about facility pumps, chillers or air handlers.
- “A liquid loop is passive once it is filled.” Active circulation and control electronics normally require power, and heat still has to reach a heat sink.
- “Redundant means independent.” Two pumps on one panel or two CDUs sharing one facility loop can fail together.
- “The system has a known number of minutes before overheating.” Thermal behavior is installation-specific; the reviewed guidance provides no universal time estimate.
- “Power restoration means immediate full load.” Pumps, valves, chillers and controls may need a staged restart and verification period.
- “One vendor’s valve behavior is standard.” Valve positions and backup-CDU switchover logic are design-specific. Chilldyne, for example, describes behavior for its own system that should not be generalized to other installations.
What the evidence supports
ASHRAE’s liquid-cooling guidance, manufacturer documentation from Trane and Eaton, and Vertiv’s design guidance agree on the central operational point: outage behavior is determined by architecture, backup power, redundancy and controls. They do not establish a universal failure rate, temperature trajectory or time-to-overheat figure. A site’s emergency plan must therefore be based on tested equipment data and documented operating limits, not on a generic number.
The practical test is straightforward: trace every component required to move heat away from the server, prove that each has a backed-up or redundant path, verify that operators receive the right alarms, and rehearse the controlled-load-reduction sequence before an outage occurs.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
