Data center humidity control and the environmental envelope field guide

Humidity control and the environmental envelope

Humidity control in a data center keeps the moisture in the air at the IT equipment inlet inside a band the gear can live with, and it is the second half of a job most people only think of as temperature. The equipment cares about two things in the air it breathes: how hot it is and how much water it carries. Hold the temperature and ignore the moisture and you still get failures, just slower ones.

The envelope is the box those two limits draw. ASHRAE calls it the thermal envelope, and it has a temperature axis and a humidity axis. The temperature side gets all the attention because a hot rack throttles or trips in minutes. The humidity side moves slower, so it gets neglected, and the bill for neglecting it shows up as static and dropped boards on the dry end or as condensation and corrosion on the wet end.

Two-sided is the whole idea. Too dry and you raise the odds of an electrostatic discharge that zaps a board during a hardware swap. Too wet and you risk water condensing on a cold surface, or hygroscopic dust pulling moisture onto a circuit and corroding it. The control problem is keeping the air between those two walls without burning energy bouncing off either one. The temperature axis, the airflow that delivers it, and the CRAC and CRAH units that do the work are covered in the cooling-and-airflow guides. This guide is the moisture axis.

What humidity should a data center run at?

A data center should hold the air at the server inlet inside the ASHRAE TC 9.9 humidity range, commonly a dew point of roughly -9 to 15 C with a 60 percent relative humidity ceiling on the recommended envelope, but the equipment's rated class and the current edition of the guidelines set the real limit. ASHRAE Technical Committee 9.9 publishes the Thermal Guidelines for Data Processing Environments, and the humidity limits live in the same envelope as the temperature ones.

The recommended envelope is the band for long-term reliability. Outside it but inside the allowable envelope, the gear still runs and the manufacturer still warrants it. You just spend some reliability margin to sit there. The allowable humidity is wider: the low end is bounded at the more-moist of a -12 C dew point or 8 percent RH, and the high end runs from about 80 percent RH on a class A1 or A2 device up toward 85 and 90 percent on A3 and A4, with a dew point cap near 17 C on A1 and 21 C on A2 to keep water off the cold surfaces.

Do not memorize one number and design to it. The recommended band is where you aim. The allowable band is how far you can drift on a bad day or to win free-cooling hours. Confirm both against the edition the design referenced and the class of the actual gear before you set anything.

Why control to dew point instead of relative humidity?

Control to dew point because relative humidity changes with temperature while the actual water in the air does not. Relative humidity is a percentage of how much moisture the air could hold at its current temperature. Warm the same air and its RH falls even though not one molecule of water left. Cool it and the RH climbs. So an RH reading without a temperature next to it tells you almost nothing about how much water is really in the room.

Dew point is the temperature at which that air would start to condense, and it is a direct measure of absolute moisture. It does not move when the air warms or cools. Two sensors reading 50 percent RH at different temperatures are looking at two completely different amounts of water. Two sensors reading the same dew point are looking at the same water, which is exactly what you want when units around a hall have to agree.

This is the reason the industry moved to dew point control. A hall controlled on RH chases a moving target, because every degree of temperature swing shifts the reading, and units in different parts of the room read different temperatures and therefore different RH from identical air. Control on dew point and the disagreement disappears, because dew point is the same number everywhere the air is the same. The relative humidity is then free to float across a wide range as the air warms and cools through the room, which is fine, because the gear cares about the moisture, not the percentage.

The wider envelope and the dew point band you set

For years rooms were pinned in a narrow RH band, often something like 45 to 55 percent, with humidifiers and dehumidifiers working constantly to hold it. That practice is gone, and good riddance, because it burned enormous energy for a reliability gain that mostly was not there. The allowable envelopes widened, the low RH limit dropped from 25 percent to 8 percent, and the control metric moved to dew point over a wide band.

The payoff is energy and free-cooling hours. A wide dew point band means the cooling plant does not have to add or pull moisture every time the outside air shifts, and an economizer or free-cooling design can pull in outside air across far more of the year without tripping a humidity limit. The free-cooling and economizer strategy lives in the cooling-and-airflow overview. The humidity band is what decides how many of those hours you actually get to use.

The practical setpoint strategy is a wide dead band on dew point and a hands-off attitude toward RH. Pick a dew point target inside the recommended range, give it a generous band on either side, and let the room float within it. The moment a unit acts on a small swing, you are back to chasing. Set the band wide enough that the equipment only humidifies or dehumidifies when the air genuinely leaves the envelope, not when a sensor twitches.

Is low humidity really an ESD risk?

Low humidity raises electrostatic discharge risk, but less than the old 45 percent RH habit assumed, which is why ASHRAE lowered the allowable floor to 8 percent RH. The fear was real in principle. Dry air lets static charge build on a body or a cart, and a discharge into an exposed board during a hardware swap can damage it. The reassessment was about how much that risk actually rises as the air gets drier when crews follow standard grounding practice.

ASHRAE-funded testing found the probability of a damaging discharge climbs only slightly as RH drops from 25 percent toward 8 percent, provided the usual ESD controls are in place. The conclusion the industry took from it is that dry air alone is not the hazard people treated it as, as long as the people and the procedures handle the charge.

So the dry side is managed with procedure, not by soaking the room. Grounded wrist straps when handling exposed electronics, ESD-rated flooring and footwear, and dissipative work surfaces do more than a humidifier ever did. The static control on a raised or access floor is its own test scope, covered with the airflow work. Bound the dry end at the allowable limit, keep the ESD program real, and you do not need to spend humidification energy defending against a risk the grounding already handles. Where the procedures are weak, the humidity floor matters more, so tighten the procedures first.

Too humid: condensation and corrosion

The wet side is the one that actually destroys hardware, and it does it two ways: condensation and corrosion. Push the dew point too high and any surface in the room colder than that dew point grows liquid water. Chilled water pipe, a cold coil, the underside of a raised floor near a supply, all of them become places water can form, and water on or inside energized electronics is a short waiting to happen.

Corrosion is the slower wet-side failure. High humidity on its own accelerates the chemistry, and combined with hygroscopic dust it gets worse, because that dust pulls moisture out of the air and holds a damp, conductive film against the board. Add corrosive gases to the air and humidity becomes the accelerant that turns trace pollution into pitted contacts and failed silver and copper traces. That gas-plus-moisture mechanism has its own section below.

This is why the high end of the allowable envelope carries a dew point cap, not just an RH percentage. A class A1 device caps around a 17 C dew point and A2 around 21 C, set specifically so the cold surfaces in the room stay above the dew point and never sweat. The number that governs condensation is the dew point against the coldest surface, not the room RH. If the chilled water is colder than the room dew point, something is going to get wet.

Condensation on the cooling coil

Every cooling coil running below the room's dew point condenses water out of the air passing through it, whether you wanted dehumidification or not. That is latent cooling, and it is free physics working against you. A CRAH coil fed with chilled water in the low 40s F is well below any normal room dew point, so it strips moisture continuously while it does its sensible job of removing heat.

The waste comes in two stages. First, the unit spends energy condensing water you did not need to remove. Then, because the room is now drier than the setpoint, a humidifier somewhere adds that moisture back, spending more energy to undo the dehumidification the coil just did. You are paying twice to hold the dew point flat.

The fix is to run the chilled water warmer. Raise the supply water temperature so the coil surface stays at or above the room dew point, and the coil does sensible cooling only, no latent. This is one of the biggest reasons modern designs push chilled water temperatures up. Warmer water means the coil stops dehumidifying for free, the humidifiers stop fighting it, and the chiller runs more efficiently on top of it. The chilled water plant and the supply temperature choices are covered in the cooling-and-airflow overview. The humidity consequence is that a coil run needlessly cold makes its own humidity problem.

The humidity fight between units

The single most common humidity waste in a data center is two cooling units fighting, one humidifying while another dehumidifies in the same hall, at the same time, holding the moisture flat while both burn energy. It is almost never a hardware fault. It is a controls and coordination failure.

The mechanism is simple once you see it. Each unit reads humidity at its own return air and acts alone. A unit working hard on sensible load runs a cold coil, so it is dehumidifying as a side effect, and its return reads dry, so it calls for humidification. A unit across the room reading a slightly different return calls to dehumidify. Now one is boiling water in while the other condenses it out, the net moisture barely moves, and nothing logs a fault. The CRAC and CRAH airflow guide walks the unit-level version of this. The room-level fix is the same three moves.

Control to dew point, not RH, so the units stop disagreeing over temperature-shifted readings. Put the fleet on a shared humidity reference or a group controller so they act as one system instead of a dozen independent ones. And widen the dead band so small swings do not trigger anyone. Do those three and the fight ends. Skip them and you can run a hall for years quietly paying for humidification and dehumidification that cancel out.

Humidification methods: steam and adiabatic

When the air genuinely needs moisture added, the choice of humidifier is mostly an energy choice, and it splits into two families. Isothermal humidifiers make steam: an electrode or resistive canister boils water, or an infrared lamp evaporates it off an open pan. They add water vapor without cooling the air, but boiling water costs a lot of electricity. Adiabatic humidifiers evaporate water into the airstream without heat: ultrasonic foggers, high-pressure spray nozzles, and wetted media that air passes through.

The energy difference is large. Steam humidification spends electricity to boil water. Adiabatic humidification uses a fraction of that, and because evaporation absorbs heat, it cools the air as it humidifies, which helps in a building that is trying to reject heat anyway. Roughly every pound of water evaporated adiabatically pulls about 1000 BTU of heat out of the air. That is why adiabatic humidification pairs naturally with free cooling and economizer operation, where you are already moving outside air and want the evaporative cooling as a bonus.

The catch on the adiabatic side is water quality and hygiene. Spraying water into the air means the water has to be clean and treated, or you put minerals and biological growth into the room. Steam sidesteps that, because boiling kills organisms and leaves minerals in the canister. Pick the method against the design's energy target and water treatment, and confirm the water source and treatment are part of the scope, not an afterthought.

Dehumidification and the latent load

Removing moisture from the air is the other direction, and most of the time a data center does it without any dedicated equipment, because the cooling coils already condense water whenever they run below the dew point. That is usually enough and often too much, which is the coil-condensation problem above. Dedicated dehumidification is the exception, not the rule, in a well-run hall.

Where a real dehumidification load shows up, it is usually coming in from outside. An economizer or makeup-air system pulling in humid outdoor air on a wet day carries a latent load the room has to remove, and in a muggy climate that load can be significant. The room's own people and infiltration add a little, but the dominant moisture source is almost always the air you bring in from outside.

The honest way to handle it is to condition the outside air before it enters the hall rather than dumping a humidity excursion into the room and chasing it with the CRAH fleet. A makeup-air unit that dries incoming air to the room dew point keeps the white-space units out of the dehumidification business. Letting raw outdoor humidity into the hall and asking the cooling units to claw it back is how you reignite the fighting-units problem on a humid afternoon.

Outside air and the economizer humidity load

Outside air is where most of a data center's humidity trouble comes from, in both directions. An air-side economizer saves a lot of cooling energy by pulling in cool outdoor air instead of running the chillers, but that air arrives at whatever dew point the weather hands you. A cold dry winter day brings air that drops the room dew point and can call for humidification. A warm humid day brings air that raises it and calls for dehumidification.

This is the tension at the center of free cooling. The colder and drier the outside air, the more free-cooling hours you get and the more humidification you may have to add to keep the room off the dry limit. The economizer saved chiller energy. The humidifier spent some of it back. Whether you come out ahead depends on the climate, the width of your dew point band, and the humidification method, which is the case for a wide band and for adiabatic humidification that costs little.

The makeup-air and economizer controls are part of the cooling design, covered in the overview guide. The humidity job is to size and sequence the outside-air conditioning so the economizer can open across as many hours as possible without throwing the room out of its envelope. A wide dew point band is what buys those hours. A narrow band closes the economizer early and gives the savings back.

Where do humidity sensors go in a data center?

Put the humidity and temperature sensors at the IT equipment inlet, in the cold aisle, not at the cooling unit return. The envelope is defined at the air the gear breathes, so that is where it has to be measured. A sensor in the return reads air that has already picked up heat and mixed across the room, which tells you about the unit, not about the rack.

Placement and quantity both matter. One sensor per hall does not see the spread across a room where the cold aisle at one end runs different from the other. Sensors distributed across the cold aisles, at the height range of the equipment intakes, give the picture the envelope is actually about. Top of rack and bottom of rack can read differently, so a single mid-height probe can miss a problem at the top where the warm air collects.

The other half is calibration. A humidity sensor drifts, and an uncalibrated one is how the fighting-units problem starts, because two drifted sensors disagree about air that is actually identical. Calibrate the sensors on a schedule and trust a calibrated dew point reading over a raw RH percentage. The readings feed the building management system and the DCIM or environmental monitoring platform, which is where you watch the whole fleet at once and catch units working against each other. Controlling and trending to the inlet, on calibrated sensors, is the foundation the rest of the humidity strategy sits on.

Controlling to the rack intake, not the return

Measuring at the inlet is one thing. Controlling the units off the inlet is the move that actually changes behavior. A unit that controls humidity off its own return is reacting to mixed, warmed air and will fight its neighbors. A unit controlling off a cold-aisle inlet reference, ideally a shared one, is holding the condition that the envelope is written against.

The same logic runs through the temperature and airflow side, where the whole point is that the cooling is judged at the inlet and not at the unit discharge or the return. The CRAC and CRAH airflow guide makes that case for temperature and air delivery. For humidity it is identical. The number that governs is the dew point at the rack inlet, so that is the number the control loop should chase.

The practical setup is a shared cold-aisle reference feeding a group controller, with the individual units trimming to hold the common dew point rather than each defending its own return reading. This is the same architecture that ends the fighting-units waste, which is not a coincidence. Control to the inlet, share the reference, and the room behaves as one system instead of a dozen arguments.

Gaseous contamination and humidity

Humidity plus corrosive gas is the mechanism that quietly kills electronics in dirty-air locations, and it is the reason the high humidity limit is not just about condensation. On its own, moderate humidity in a clean room does little harm. Add gaseous contaminants, hydrogen sulfide, sulfur dioxide, chlorine compounds, the kind of air you get near heavy industry, agriculture, or certain urban environments, and humidity becomes the accelerant that drives those gases to attack the copper and silver in the hardware.

ASHRAE addresses this in its work on gaseous and particulate contamination, separate from the thermal guidelines but tied to them through humidity. The standard tool is corrosion coupons: small strips of copper and silver placed in the room and measured over time for how fast they corrode, which classifies the air's severity. A site that comes back reactive needs better filtration, gas-phase filtration in the worst cases, and a tighter handle on humidity, because the same moisture that was harmless in clean air is now feeding the corrosion.

The field point is that humidity does not act alone on the wet side. In a clean room you can run the upper allowable band without much worry. In a contaminated one, the same humidity is dangerous, so the air quality and the humidity limit have to be set together. Put coupons out, read them, and let the result, not an assumption that the air is clean, set how hard you hold the upper humidity limit.

Liquid cooling and the room

As racks move to direct-to-chip and immersion liquid cooling, the air humidity matters less for the chips and not at all for the part of the load on the liquid loop, but it does not go to zero for the room. The silicon being cooled by liquid does not breathe room air, so its dew point exposure is whatever happens at the liquid loop, not in the hall. That shifts the humidity concern but does not remove it.

The room still has air-cooled gear in it: network equipment, power distribution, the air-cooled remainder of hybrid racks, and people. That equipment still has an envelope, and the room still has cold surfaces that can sweat. A direct-to-chip loop running technical-cooling water near room temperature has its own condensation concern. If that loop ever runs below the room dew point, it sweats inside the rack, which is exactly the place you do not want water. So the room dew point still has to stay below the coldest liquid-loop surface.

For now, most halls are mixed, so the humidity envelope still applies to the air side even as the heat increasingly leaves by water. The liquid-cooling architecture is covered in the cooling-and-airflow overview. The humidity takeaway is that liquid cooling changes where the dew point matters, from the server inlet toward the liquid loop surfaces, without letting you stop controlling it.

WUE: what humidification water costs

Adiabatic humidification and evaporative cooling save energy by spending water, and that trade has a metric: water usage effectiveness, WUE, the liters of water a facility uses per kilowatt-hour of IT energy. Every gallon evaporated into the air to humidify or to cool is a gallon counted against WUE, and in a water-stressed region that number can matter as much as the energy it saved.

The tension is direct. Adiabatic humidification and evaporative cooling cut electricity by using the heat-absorbing power of evaporating water, which is good for PUE and for free-cooling hours. The same water shows up on the WUE side of the ledger. A design that optimizes PUE with heavy evaporative use can post a poor WUE, and in some climates and jurisdictions the water is the binding constraint, not the power.

There is no single right answer, because it depends on what is scarce where you are building. In a cool, water-rich climate, leaning on evaporative humidification and cooling is an easy win. In a hot, dry, water-stressed one, the WUE cost can push the design back toward steam humidification, or toward simply running a wider dew point band so you humidify less in the first place. Know which resource is tight on the site before you pick the humidification method, because the two metrics pull against each other.

Matching the control to the equipment class

The humidity limits you hold are set by the class of the gear in the room, A1 through A4, not by a single industry number. ASHRAE assigns IT equipment to those classes, and each class carries its own allowable temperature and humidity envelope. A1 is the tightest, typically older enterprise servers and storage with the narrowest humidity tolerance. A2, A3, and A4 progressively widen the allowable band, with A3 and A4 built to tolerate the widest swings.

The binding constraint is the most sensitive device in the room. A hall full of A3-rated gear with one rack of older A1 storage is an A1 hall for the purpose of setting the humidity limits, because the limit has to protect the weakest equipment, not the average. This is the trap in reading one ASHRAE recommended number and applying it everywhere. The recommended band is common to all classes, but the allowable band, the room you have to maneuver in, depends on the actual class mix.

Get the class list before you set the band. Pull it from the equipment specifications and the manufacturer documentation, confirm the edition of the guidelines the design referenced, and set the humidity limits to the most restrictive class present. Then the wide-band, dew-point strategy operates inside limits that genuinely protect the gear instead of limits you assumed.

Commissioning the humidity control

Humidity control gets tested at turnover the same way the rest of the mechanical plant does, and it gets skipped more often, because it is invisible until it fails. The components come first: the humidifiers and any dehumidification confirmed installed and operating, the water supply and treatment for adiabatic units proven, the inlet humidity sensors installed in the right place and calibrated against a reference.

Then the sequences. Force a humidity excursion, drive the room dry and watch the humidifiers respond, drive it humid and watch the dehumidification or the coil response, and confirm the whole fleet acts as one system instead of individual units chasing their own returns. The test that earns its keep is watching every unit at once during the excursion, because the only way to catch two units fighting is to see them both. If one humidifies while another dehumidifies at steady state, the control setup failed the test no matter what the room average reads.

Last, prove the band. Confirm the dew point dead band is set wide enough that small swings do not trigger action, and that the room can float across the band without anyone humidifying or dehumidifying needlessly. Document the setpoints, the band, the sensor calibration, and the excursion results, because the next operator inherits whatever you signed off and needs to know what normal looks like.

What to document

The humidity strategy that is not written down gets undone by the next technician who decides the room feels dry and tightens a setpoint. Record the design decisions and the as-commissioned conditions so the band survives staff turnover and the next humidity complaint can be checked against what was actually set.

Common mistakes

• Controlling to relative humidity instead of dew point, so units chase temperature-shifted readings and disagree about identical air.
• Running humidifiers and dehumidifiers in the same hall at once, the fighting-units waste, instead of a shared reference and a wide band.
• Holding a tight, narrow humidity band like the old 45 to 55 percent RH practice, burning energy for reliability margin the gear does not need.
• Controlling and sensing at the unit return instead of the IT equipment inlet where the envelope is defined.
• Running the chilled water or coil needlessly cold, so it condenses moisture the room then pays a humidifier to add back.
• Treating the air as clean and ignoring gaseous contamination, so the upper humidity limit is set without coupons or filtration.
• Trusting uncalibrated humidity sensors, which drift apart and make the room think it is dry where it is not.
• Setting the band off one ASHRAE number without checking the actual equipment class or the edition the design referenced.

Field checklist

Standards and references

The central reference is ASHRAE Technical Committee 9.9 and its Thermal Guidelines for Data Processing Environments, which set the recommended and allowable temperature and humidity envelopes and the A1 through A4 equipment classes. The humidity limits, the dew point bounds, and the low-RH floor that dropped to 8 percent all come from there. Confirm the edition the design referenced, because the envelopes and the class definitions have shifted across editions and the current one governs.

For the wet-side risk beyond condensation, ASHRAE's work on gaseous and particulate contamination is the companion reference, the source for the corrosion-coupon approach and the air-quality classes that decide how hard to hold the upper humidity limit. The energy side of the building, including the economizer and free-cooling design that drives much of the outside-air humidity load, is bounded by ASHRAE Standard 90.4 for data centers, with 90.1 and 62.1 covering the parts of the building outside the white space.

Above all of it sits the IT equipment's own specification. The manufacturer's stated environmental limits can be tighter than any ASHRAE band, and where they are, the listing governs and the project specification controls the call. Cite the standard that actually sets the point, hedge the ranges to the current ASHRAE edition and the gear, and let the equipment spec win when it is stricter.

Units, terms, and conversions

Humidity shows up in several units across a drawing set, a controls submittal, and a manufacturer sheet, and they do not all measure the same thing. Keeping them straight is half of not misreading a setpoint.

Dew point is given in degrees C or F and is a direct measure of absolute moisture. Relative humidity is a percentage tied to the current air temperature. Absolute or specific humidity is the actual mass of water per mass of dry air, in grams per kilogram or grains per pound. The psychrometric chart is the tool that ties temperature, RH, and dew point together, and reading one is the skill behind understanding why an RH number alone is incomplete.

Dew point

The temperature at which air begins to condense, a direct measure of absolute moisture

Relative humidity (RH)

Moisture as a percentage of what the air could hold at its current temperature

Absolute / specific humidity

Actual mass of water per mass of dry air, independent of temperature

Thermal envelope

The temperature and humidity box the IT equipment is held within

Latent vs sensible

Latent cooling removes moisture; sensible cooling removes heat without changing moisture

Equipment class (A1-A4)

ASHRAE rating that sets a device's allowable temperature and humidity band

WUE

Water usage effectiveness, liters of facility water per kWh of IT energy

Corrosion coupon

A copper or silver strip used to measure air corrosivity over time

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What underfloor static pressure should a data center plenum hold?
• What is bypass airflow in a data center?
• What is the ASHRAE recommended temperature for a data center?
• What humidity should a data center be?
• What is a CDU in a data center?