The Cooling Risk Outside the Pipe
By Brenda Pak
A seal keeps coolant in. It cannot stop moisture in the room from settling on cold metal. Before lowering a setpoint, check what happens outside the loop.
You find water on a cooling connection. The first suspect is the seal.
That is a reasonable place to look. But water has another way to reach the hardware: it can arrive from the air. Moisture condenses on a cold surface even when the coolant circuit is perfectly sealed.
Now consider the setpoint conversation. The GPUs are running warmer, and someone suggests colder coolant. That may improve thermal headroom at the chips. It can also take an exposed connector, valve or pipe closer to the point where moisture starts collecting on it.
Water near a rack still needs the site's response procedure. Its appearance cannot tell you where it came from. Keep both sources in the investigation, and bring one question to the setpoint review: what air will the coldest exposed surface meet?
Same humidity. A different outcome.
“The room is at 50% humidity” sounds reassuring. It leaves out half the calculation.
Dew point is the temperature at which water vapour begins to condense as air cools. To calculate it from relative humidity, you also need air temperature. The National Physical Laboratory sets out the relationship.
Take a surface held at 18°C. Compare two air conditions, both at 50% relative humidity:
- At 24°C, the calculated dew point is 12.9°C. The surface is 5.1°C above it.
- At 30°C, the calculated dew point is 18.4°C. The same surface is now 0.4°C below it.
Nothing happened to the seal. The metal is equally cold. Yet the second combination puts the surface below dew point.
Compare two room conditions
How the margin disappears
Follow the dew point toward the unchanged metal temperature.
Same surface. Same relative humidity. Two different room conditions.
Try your own temperature comparison
The surface is 5.1°C above dew point. The required design allowance is not included.
See how the dew point was calculated
Two different dew points.
Changing the room conditions can erase the margin, even when the metal stays just as cold.
The comparison uses NPL's Magnus relationship, with air temperature T in °C and relative humidity RH in percent.
γ = ln(RH / 100) + 17.62T / (243.12 + T)Dew point = 243.12γ / (17.62 − γ)The controls cover 20°C to 35°C air temperature. The result estimates dew point, not a permitted operating setpoint.
This comparison holds relative humidity at 50%; it does not represent heating a sealed parcel of air with unchanged moisture content. It separates two room conditions so the missing variable is easy to see.
Humidity can change the answer too. At 24°C and 70% relative humidity, dew point is about 18.2°C. That also puts an 18°C surface below the threshold. Try those values in the comparison above. The useful pair is air temperature and relative humidity, measured together.
Put the measurement on the right side of the wall
A supply sensor tells you the fluid temperature at that location. Condensation happens on a surface, in contact with local air. Those are different measurements.
Start with the cold surfaces people can actually point to: connectors, valve bodies and exposed sections of pipe. Which ones are insulated? Where does the insulation end? Which air-temperature and humidity readings represent those locations?
Find the measurement
Three readings. Two sides of the wall.
The fluid is inside. Condensation happens outside. Choose a reading to see the question it answers.
Surface temperature: the other half of the comparison
Measure the cold surface exposed to room air. Compare it with local dew point, including the allowance required by the design. A coolant sensor alone does not establish this surface temperature.
Insulation details matter here. ASHRAE's moisture-management guidance describes the role of insulation and vapour control around cold pipes and valves. A well-insulated straight run does not answer for an exposed service connection beside it.
Sensor placement matters too. Vaisala's humidity guidance explains why the measurement must represent the air of interest and why temperature differences affect the result. Before reusing the nearest room sensor, establish what it represents.
An uninsulated valve beside an insulated pipe is an easy detail to miss on a dashboard. Walk the accessible route with the design drawings. Match each relevant cold surface to the air measurements that represent it.
The restart deserves its own answer
The steady running trend may be the most reassuring screenshot in the review. It is also only one operating state.
Ask what happens at initial fill, at a cold restart and in the maintenance mode that changes the usual flow path. A comfortable margin during normal operation does not establish the margin in those conditions.
The Open Compute Project's liquid distribution guidance, page 9, flags transition and maintenance conditions in which a technology cooling loop can fall below dew point. It also describes the CDU's role in separating facility water from the circuit serving the IT equipment.
A liquid-to-liquid heat exchanger separates the fluids. It does not, by itself, prove that every operating mode maintains the required temperature margin. That answer belongs in the installed system's control sequence and commissioning evidence.
Take the review beyond the steady trend
Three states. Three checks.
Running
Which exposed surface has the smallest margin in the expected room conditions?
Starting again
What keeps the loop within its permitted conditions during the coldest allowed start?
In maintenance
Does a changed flow path or control sequence change the surfaces that need checking?
Check those cases through the approved commissioning process. Do not create a condensation event on a live rack to test the theory.
A tenth of a degree is an answer, not an allowance
A surface at 20.0°C is above a calculated dew point of 19.9°C. By a tenth of a degree. That positive result is arithmetic, not a reason to approve the setting.
Sensor uncertainty, changes between readings and control response all matter. The required allowance must come from the design and equipment requirements, rather than a universal number borrowed from an article.
Look closely at the word non-condensing in equipment specifications. For example, it appears in Lenovo's environmental limits for the GB300 NVL72. A stated humidity range does not cancel that condition. Use the documentation for the installed equipment, not another rack's limits.
Keep the load question and the condensation question together. A change needs to serve the IT equipment while staying inside the cooling system's permitted operating conditions.
Bring this to the setpoint review
Make the proposed change easy for the next person to review. One record should connect the setting, the measurements and the response if the margin is lost.
Six items to put beside the proposed setpoint
| The evidence | The decision it supports |
|---|---|
| Proposed setpoint and permitted equipment range | Is this change allowed for the installed hardware and coolant? |
| Local air temperature, humidity and sensor locations | Which dew point applies to the exposed surfaces? |
| Coldest expected surfaces and insulation condition | Where is the smallest temperature margin? |
| Measurement uncertainty and the specified allowance | Is that margin large enough to rely on? |
| Start, restart and maintenance sequences | Does the reasoning still hold outside normal running? |
| Alarm behaviour and assigned response | Who acts when the required margin is lost? |
Use this as a discussion checklist alongside the site's design criteria and procedures. It intentionally gives no universal operating temperature.
First, make sure colder is the right fix
A hotter GPU can mean more work, a warmer supply or reduced rack flow. Those are different investigations, as we explain in The GPUs Are Hotter. Is the Cooling Loop to Blame?.
If the problem is elsewhere, reducing the supply setpoint can leave it unresolved while taking away condensation margin.
Establish why the GPU is hotter first. If a colder supply is still justified, check the exposed surfaces and the local air before applying it.
The seal answers whether coolant can get out. The dew-point comparison answers whether water can arrive from somewhere else. A reliable setpoint decision needs both.
References
- National Physical Laboratory: Converting between dew point and relative humidity. Dew point definition and Magnus relationship used in the comparison.
- ASHRAE Handbook: Moisture Management in Buildings. General cold-surface, insulation and vapour-control principles.
- Vaisala: How to Choose the Right Instrument for Measuring Humidity in Your Application. Representative placement and temperature effects.
- Open Compute Project: Data Center Liquid Distribution Guidance and Reference Designs, revision 1.0. Page 9, CDU planning and transition conditions.
- Lenovo: GB300 NVL72 environmental specifications. Equipment-specific non-condensing requirement.
Subscribe to updates
Get the latest engineering perspectives sent straight to your inbox.