TL;DR: Forty racks at 120 kW create a 4.8 MW IT heat load. If all of that heat enters a water loop with a 10 K temperature rise, the loop needs about 114.8 kg/s, or 6,890 L/min. A real design must split liquid and air loads, use CDU capacity at the actual operating temperatures, and prove that the piping and controls can maintain cooling after the required failure. Temperature, coolant chemistry and shutdown behavior come from the supported equipment design, not a universal rule of thumb.
How to approach it
Establish the heat load and liquid capture fraction. Derive the flow, then check hydraulic limits, heat-exchanger performance and redundancy. Finish with coolant compatibility, protective controls and commissioning evidence. Keep the facility loop separate from the coolant that reaches the servers.
A strong answer
A rise of 10 K (kelvin) is the same size as a 10 °C rise: the water leaves 10 degrees warmer than it arrives. Suppose a hall will contain forty racks rated for a 120 kW design load each. The building has an existing 12°C chilled-water supply. Before choosing whether to reuse it, obtain the server's allowed coolant temperatures, flow range, pressure drop, fluid specification and residual air-cooling requirement. The existing plant may be useful, but its supply temperature should not automatically become the server-loop setpoint.
Convert heat into flow
At steady state, almost all electrical power entering the IT equipment becomes heat. The simple calculation below assumes the entire 4.8 MW enters a water loop. It uses a heat capacity of 4,180 J/(kg·K) and density of approximately one kilogram per litre.
Heat: 40 × 120 kW = 4,800 kW = 4.8 MW
Q = mass flow × specific heat × temperature rise
At a 10 K rise:
mass flow = 4,800,000 / (4,180 × 10) = 114.83 kg/s
volume flow ≈ 114.83 × 60 = 6,890 L/min
per rack ≈ 172.25 L/min
At 5 K: about 13,780 L/min for the same heat
At 20 K: about 3,445 L/min for the same heat
sanity: halving the temperature rise doubles required flow at fixed heat and fluid.
This is an energy balance, not a complete pipe or cold-plate design. The system must also meet minimum device flow, maximum pressure, pump head and coolant-temperature limits. Glycol mixtures change heat capacity, density and viscosity. A higher temperature rise can produce hotter return coolant and higher device temperatures; it does not independently determine the supply temperature.
If an illustrative 90% of the IT heat enters liquid, the liquid duty is 4.32 MW and the remaining air load is 480 kW. At the same water properties and 10 K rise, liquid flow is about 6,201 L/min. Add relevant pump heat and losses at their actual boundaries. Do not leave the residual air load out of the room design.
Size CDUs at the design point
A coolant distribution unit separates the controlled server loop from the facility loop through a heat exchanger. Direct-to-Chip Liquid Cooling and CDUs shows these two paths. CDU capacity depends on fluid temperatures, flow and the heat exchanger's approach temperature: the temperature difference needed to transfer the heat. A nominal “500 kW” label alone is insufficient. For a concrete manufacturer example, the Vertiv CDU 600 planning guide specifies capacity at a stated approach temperature and describes the separate fluid circuits.
Suppose a chosen unit can actually reject 500 kW at this design point. With the full 4.8 MW assigned to liquid, ten running units provide 5 MW. Eleven connected units could provide capacity N+1: after losing one, ten remain. That leaves only 200 kW, or about 4.17%, above the load after failure, so derating or a larger design margin may require more units.
The arithmetic does not prove the redundancy. A spare CDU must have a valid hydraulic path, sufficient pump head, working controls and power to serve the affected racks. An isolated spare beside the wrong group cannot replace a failed unit. Shared headers, isolation valves, primary pumps, controls and upstream heat rejection also create failure domains.
Assume each CDU can reject 500 kW at the specified operating conditions.
Capacity check
- 11 installed; 10 after one fails
10 × 500 kW = 5 MW, leaving 200 kW above the 4.8 MW example load.
Delivery check
- Headers, pumps and controls must work
Verify surviving units can deliver required flow to every affected rack without exceeding pressure or temperature limits.
Choose the grouping and maintenance policy explicitly. If the requirement includes another failure while a unit is serviced, simple N+1 may be insufficient. Two complete independently capable paths are closer to a 2N design than to “one extra unit”; their shared dependencies still need inspection.
Choose temperatures with the whole system in view
Warmer facility water can reduce chiller use when ambient conditions and equipment limits permit. A dry cooler needs a temperature difference to the outdoor air; ambient merely below the desired supply temperature does not guarantee adequate capacity. Account for approach temperatures across both the outdoor heat rejection equipment and the CDU.
Compare a warm-water design with a controlled connection to the existing chilled plant over the site's climate and operating range. Keep server coolant above the required dew-point margin, within the server's inlet specification and compatible with all connected devices. A 12°C primary loop can transfer heat through a CDU to a warmer secondary loop; it is not automatically an invalid design. Capital cost, efficiency, redundancy and commissioning schedule decide whether reuse is worthwhile.
Specify the fluid and the failure response
Use the server, cold-plate and CDU suppliers' compatible coolant requirements. Specify filtration, cleanliness, corrosion protection, materials, seals, sampling and maintenance as a coordinated system. There is no universal 50-micron filter or conductivity threshold for every loop. Conductivity can be a useful monitored property, but its acceptable range depends on the approved coolant; a high reading alone is not proof of corrosion.
Monitor supply and return temperatures, flow, pressure differential, leaks and equipment temperatures. Relate alarms to the action they require. A partial blockage, pump failure and leak may need different isolation and shutdown responses. A software power cap may reduce heat, but it is not a guaranteed substitute for an equipment-approved protective interlock.
Thermal ride-through depends on fluid inventory, thermal mass, load, flow paths and hardware protection. Do not promise a fixed “tens of seconds” window or assume a checkpoint can finish after cooling stops. Validate automatic workload rejection, load reduction and shutdown sequencing against the supported system. Thermal, Power and Cooling Events explains the operational side.
Commission with an approved test plan: verify pressure/leak integrity, fluid quality, balancing, design-point heat removal, sensor accuracy and the required loss-of-unit and maintenance cases. Use staged loads or appropriate test equipment and vendor procedures; an uncontrolled flow interruption on valuable hardware is not an acceptance test. Capture the actual margins and recovery behavior.
If the racks or liquid capture fraction change, redo both liquid and air calculations. A rear-door heat exchanger may suit an air-cooled rack, but it cannot replace cold-plate cooling required by a server, and there is no universal 70 kW ceiling across all rear-door products.
What interviewers probe next
- “Why choose a 10 K rise?” It is an explicit example. The final rise must satisfy device flow, temperature and hydraulic requirements.
- “What makes the spare useful?” A tested path and adequate capacity through every shared component, not the unit count alone.
- “Can the chilled loop be reused?” Possibly, through a compatible controlled design; compare lifecycle cost and operating limits.
Common mistakes
- Assigning 100% of heat to liquid without checking the capture fraction.
- Treating CDU nameplate capacity as independent of temperatures and flow.
- Calling an unreachable spare N+1 protection.
- Assuming an alert leaves enough time for a human or a checkpoint.
Key takeaways
- The full-liquid example needs about 6,890 L/min at 4.8 MW and a 10 K rise.
- At 90% liquid capture, approximately 480 kW still needs air-side removal.
- Validate capacity and flow after failure, then verify the controls and maintenance path under the same requirement.