Datacenter
Data center thermal envelope and ASHRAE setpoints field guide
What the IT intake air is allowed to be: the ASHRAE TC 9.9 recommended and allowable ranges, raising the supply setpoint to cut cooling energy, dew-point humidity, and the inlet rule that decides all of it.
Direct answer
The data center thermal envelope is the temperature and humidity range for the air entering the IT equipment, set by ASHRAE TC 9.9. The recommended band is about 18 to 27 C (64 to 81 F), with wider allowable ranges by equipment class. Measure it at the rack inlet, not the room. The equipment manufacturer controls the limit.
Key takeaways
- ASHRAE TC 9.9 recommended intake range is about 18 to 27 C (64 to 81 F), same for classes A1 through A4.
- Measure the thermal envelope at the IT equipment intake at the rack face, never the room average or the return air.
- Allowable ranges are wider and set by class (A2 up to roughly 35 C, A3 near 40 C, A4 near 45 C) for short excursions only, not a higher steady setpoint.
- In a mixed aisle the lowest equipment class present governs the limit, and the manufacturer sets the class.
- Raise the supply setpoint only with sealed containment and inlet monitoring, walking it up in steps while watching the worst inlet, and control humidity to a wide common dew-point band.
The thermal envelope, and what it actually governs
The data center thermal envelope is the temperature and humidity range the air is allowed to be when it reaches the front of the IT equipment. ASHRAE Technical Committee 9.9 publishes it in the Thermal Guidelines for Data Processing Environments, and almost every cooling decision in a hall traces back to it. The envelope is not a single number. It is a window, with a temperature band and a moisture band, and the gear is built to run inside that window.
Two ideas make the envelope useful. The first is that the air the server cares about is the air going into it, the cold-aisle intake, not the air in the room and not the warm air coming back to the cooling unit. The second is that the window is wider than most halls actually use. Plenty of rooms run cold out of habit, sitting near the bottom of the band and paying for cooling they do not need.
Run to the envelope and you do two things at once. The equipment stays inside the conditions it was designed for, so reliability holds, and the cooling plant does less work, because warmer supply air is cheaper to make and opens more hours where the outside air can do the cooling for free. The containment that separates hot and cold air and the CRAC/CRAH airflow that delivers it are the systems that let you hold the envelope at every rack. This guide is about the target those systems are aiming at.
Why the envelope matters: too cold, too hot, wrong humidity
The envelope exists because both ends of the range cost you something, and the cost is different at each end. Too cold is the quiet one. Air below the recommended band rarely hurts the gear, so nobody notices, but the cooling plant is burning energy to hold a temperature the equipment never needed. A hall running at 18 C when the gear is happy at 24 C is paying a cooling bill for no reliability return.
Too hot is the loud one. Push the intake above what the equipment is rated for and you get throttling first, where processors slow themselves to shed heat, and a reliability hit over time as components run hotter than designed. Throttling is the tell that the cooling lost the room, and it shows up as lost compute before it shows up as a failure.
Humidity has two failure modes that pull in opposite directions. Air that is too dry raises the risk of electrostatic discharge, the static that bites a technician and can damage a component during a swap. Air that is too wet brings condensation, corrosion on contacts and boards, and over years the zinc whisker growth that flakes off galvanized surfaces and shorts things out. The envelope is the band where none of those four problems, over-cooling, overheating, static, or condensation, is the one you are fighting.
Where do you measure data center temperature?
You measure the thermal envelope at the IT equipment intake, the front of the rack in the cold aisle, not the room average and not the return air at the cooling unit. This is the rule everything else hangs on, and it is the one that gets missed. The server pulls air across its boards at the inlet, so the inlet temperature is the only one that describes what the hardware actually experiences.
A room reading and a return reading both lie to you, in opposite directions. The room average blends cold supply with warm exhaust and reports something no piece of equipment ever breathes. The return air at the CRAH is warm by design, because it has already picked up the server heat, so judging the envelope by the return tells you the gear is hot when the inlets may be fine, or hides a hot inlet behind a cool average. A unit can blow a perfect 18 C off its coil and still leave the top of a distant rack pulling 30 C, because the cold supply short-circuited back before it climbed the rack face.
The practical version is a sensor at the rack inlet, and on a tall rack more than one, because the temperature is not uniform top to bottom. Hot-aisle and cold-aisle containment exists in large part to make the inlet uniform so a single reading means something. Until the air is contained, the inlet you measure and the inlet two feet away can disagree by several degrees.
The ASHRAE equipment classes A1 through A4
ASHRAE TC 9.9 sorts IT equipment into classes by how wide an allowable range the gear can take. The data center classes commonly run A1, A2, A3, and A4, with a high-density air-cooled class H1 added in the 2021 edition. A1 is the tightest allowable range and A4 is the widest. The class is a property of the equipment, set by the manufacturer, not a setting you choose for the room.
As a rough orientation, older or more sensitive gear is built to A1, with the narrowest allowable band. Most mainstream servers shipped in recent years are rated A2, which allows intake air up to roughly 35 C. A3 and A4 push the allowable ceiling higher again, commonly cited near 40 C for A3 and 45 C for A4, which is what makes wide economizer operation and warmer halls possible without leaving the allowable range. The exact limits per class come from the current edition of the Thermal Guidelines, so confirm the numbers there and against the equipment.
The 2021 fifth-edition guidelines added an air-cooled class aimed at high-density and AI equipment, class H1. Unlike the A-classes, H1 carries a narrower recommended band, about 18 to 22 degrees C, and a tighter allowable ceiling near 25 degrees C, because the dense, high-power gear it covers needs cooler intake air to hold its thermal limits. So high density does not always mean a warmer hall: an H1 row wants colder supply air than an A2 row, not warmer. Confirm the class and its band against the current ASHRAE edition and the equipment data sheet.
The class that matters for a mixed hall is the lowest one present. If a row of A2 servers shares a containment aisle with one A1 device, the A1 limit governs that aisle, because the air is common to all of it. Knowing the class of every piece of gear in an aisle is what lets you set a setpoint with confidence instead of guessing low to be safe.
| Class | Typical use | Allowable ceiling (confirm against TC 9.9) |
|---|---|---|
| A1 | Older or more sensitive enterprise gear, tightest band | Lowest of the data center classes |
| A2 | Most mainstream servers, common default | Up to roughly 35 C |
| A3 | Equipment rated for wider operation | Up to roughly 40 C |
| A4 | Widest allowable, supports aggressive economizing | Up to roughly 45 C |
What is the difference between recommended and allowable?
The recommended range is the conservative band for long-term operation. The allowable range is the wider band the equipment will tolerate, mainly for shorter excursions. ASHRAE TC 9.9 defines both, and the difference between them is the single most useful idea in the whole envelope, because it is what lets you run warm without running scared.
The recommended temperature range is the same for classes A1 through A4: about 18 to 27 C, or 64 to 81 F. That is the band where you want the inlets to sit most of the year. The allowable range is wider and it is set by class, so an A2 device tolerates intake air well above 27 C when it has to, and an A4 device tolerates more again. The allowable band is not a free upgrade to a higher steady setpoint. It is headroom.
The way operators actually use the two: hold the inlets inside the recommended band as the normal condition, and let them ride up into the allowable band during the times that earn it, a hot afternoon on an economizer or a cooling unit out for service. The recommended band protects the long run. The allowable band absorbs the events. Treat the allowable ceiling as a place you visit, not a place you live.
| Range | Temperature | How to use it |
|---|---|---|
| Recommended (all classes A1-A4) | About 18 to 27 C (64 to 81 F) | Normal steady-state target for the inlets |
| Allowable (by class) | Wider, up to the class ceiling | Short excursions, economizer peaks, unit out for service |
Setting the supply temperature inside the envelope
The temperature setpoint that matters is the cold-aisle or supply temperature, and the target is to land the worst inlet inside the recommended band, not the average inlet. You set the cooling supply so that the hottest rack inlet in the worst spot still sits where you want it, which means the rest of the hall sits a little cooler than that. Design to the worst case, because the worst case is the rack that fails first.
Where you set it inside the band is an energy decision. A warmer supply temperature costs less to produce, and it opens more hours where an air-side or water-side economizer can carry some or all of the load without running the chillers hard. Warmer chilled water lets the cooling towers do more of the work directly. Every degree you can give back to the band, while keeping the worst inlet safe, is energy you stop spending.
The number people carry is that cooling energy moves several percent for each degree of inlet temperature, with figures around 4 to 5 percent per degree F cited in the efficiency guidance. Treat that as a planning rule, not a guarantee, because the real savings depend on the climate, the economizer, and how the fans respond. The point holds either way. Supply temperature is the biggest energy lever you set, and most halls leave it on the table by running colder than the envelope requires.
Can you raise the data center temperature to save energy?
Yes, and many data centers should, because they run colder than the gear needs. Raising the supply temperature toward the top of the ASHRAE recommended band, near 27 C at the inlet, cuts cooling energy and adds economizer hours. The catch is that you raise it with containment and monitoring in place, not by turning a dial and hoping.
The reason the dial alone is dangerous is mixing. In an open hall without containment, hot exhaust loops back into the cold aisle, so the inlet you read at one rack does not match the inlet two racks down. Raise the setpoint in that room and the worst inlet, the one you were not watching, goes over the line first. The savings are real but the risk lands on the rack you did not measure.
The modern practice is to contain the aisles, instrument the inlets so you can see the worst one, then walk the supply temperature up in steps while you watch the hot spots and the equipment for throttling. Raise it, hold it, confirm the worst inlet is still inside the band at full load, then raise it again. Done that way, the higher setpoint is a measured gain. Done blind, it is an outage waiting for a hot day. The containment field guide covers the seal that makes this safe, and the CRAC/CRAH airflow guide covers delivering the air evenly enough that the step-ups hold.
The humidity band: too dry and too wet
Humidity is the second axis of the envelope, and it has a floor and a ceiling for different reasons. Too dry and you invite electrostatic discharge, the static that builds on dry air and surfaces and can damage a component during handling or a hot swap. Too wet and you risk condensation on cold surfaces, corrosion of contacts and copper, and over time zinc whisker growth off galvanized parts that breaks loose and causes shorts.
ASHRAE TC 9.9 expresses the recommended moisture band in terms of dew point with a relative-humidity context, and the allowable band is wider. Commonly cited recommended figures put the upper moisture limit around 60 percent relative humidity with a dew-point ceiling near 15 C, and a dry-side limit set by a minimum dew point rather than a fixed low relative humidity. Confirm the exact numbers against the current edition and the equipment, because the moisture limits have moved across editions more than the temperature band has.
The reliability lesson on humidity is that the dangerous condition is usually the dry side during winter, when cold outside air is brought in and warmed, which drops its relative humidity hard. That is the season halls fight static and reach for humidification. The wet side is the corrosion and condensation problem, worse in humid climates and made worse by gaseous contamination, where ASHRAE advises holding the upper limit lower when copper and silver corrosion is a concern.
Why ASHRAE moved to dew-point limits
ASHRAE TC 9.9 frames the modern moisture envelope around dew point, not just relative humidity, and that change matters in practice. Dew point measures the actual amount of water in the air, which does not change as the air moves through the hall and warms up. Relative humidity does change, because it depends on temperature, so the same air reads one relative humidity at the cold aisle and a lower one a few degrees warmer at the inlet.
Controlling to a relative-humidity number alone makes the cooling units chase a moving target. As supply temperature rises, the relative humidity of that air falls even though no water was removed, so an RH-based controller starts adding moisture that the gear never needed. Controlling to dew point fixes the reference to the real water content and stops that chase.
The wider dew-point band the modern guidelines allow is the practical payoff. A hall that holds dew point within a wide window instead of pinning relative humidity to a tight setpoint runs the humidifiers and dehumidifiers far less. Less moisture control is less energy and less wear on the equipment that does it. If your controls still trend and alarm on relative humidity alone, that is the first thing to revisit before you blame the plant for high humidity energy.
The humidify-versus-dehumidify fight
Tight humidity control wastes energy in a way that hides on the bill, because two systems can end up working against each other. Set a narrow relative-humidity band across several cooling units and one unit can be humidifying while another, reading slightly differently a few feet away, is dehumidifying. Both run, both draw power, and the net change in the room is close to zero. This is the classic data center energy leak.
The fix is a wider band and a common reference. Widen the moisture window to what the dew-point envelope actually allows, and control all the units to the same dew-point target rather than each to its own local relative humidity. The fighting stops because there is room between the floor and the ceiling, and because the units agree on what they are aiming at.
On most days, in most climates, a hall held to a wide dew-point band needs very little active humidity control at all. The moisture the IT load and the building add tends to keep the air in range on its own. The energy you were spending was the energy of two machines arguing. Once the band is wide and shared, that argument ends, and the humidity-control energy line on the report drops to where it should have been.
The x-factor: reliability versus temperature
The fear behind running warm is that hotter air means more failures, and ASHRAE addresses it directly with the x-factor, a relative failure-rate figure indexed to a 20 C baseline. At 20 C inlet the x-factor is 1.0. Run hotter and it rises, run cooler and it falls. The number lets you reason about reliability instead of guessing.
What the x-factor shows is that the failure-rate increase with temperature is real but modest. ASHRAE's published figures put the annualized hardware failure rate at a continuous 25 C inlet somewhere in a range above the 20 C baseline, with a midpoint increase that is meaningful but far from alarming, and even continuous operation near the top of the allowable range raises the absolute failure rate by only a few percentage points on top of a base rate that is already low.
The insight that makes warmer operation defensible is time at temperature. You do not run at the ceiling all year. With an economizer, the hall runs cool for the many hours the weather is cool, which pulls the x-factor below 1.0 for those hours, and the cool hours more than offset the hot ones. The net effect across the year, weighting each hour by how long it lasts, can come out at or below the reliability of a hall held at a constant moderate temperature. The tradeoff is favorable, but it depends on the climate and the hours, so weight your own profile rather than assuming it.
Containment is what lets you raise the setpoint
You cannot safely raise the supply temperature in a hall where hot and cold air mix, so containment comes first and the higher setpoint comes second. Hot-aisle or cold-aisle containment puts a physical barrier between the cold supply and the hot exhaust, which does two things that the setpoint depends on. It stops exhaust from looping back and raising the inlet, and it makes the inlet temperature uniform across the rack face so one reading describes the whole aisle.
Without that uniformity, the gap between the average inlet and the worst inlet is your hidden risk. You set the supply to a comfortable average and the top corner of a distant rack is several degrees hotter, already near its limit. Raise the average and that corner goes over. Containment closes the gap, so the average and the worst inlet converge, and the headroom you measure is the headroom you actually have.
This is why the order is fixed. Contain the aisle, confirm the seal, verify the inlets are uniform, and only then walk the supply temperature up. Try it in the other order and the first hot day finds the rack you were not watching. The aisle containment field guide covers the leakage paths, the blanking panels, and the differential pressure that proves the seal is doing its job.
Sensors: where to put them and what to watch
The envelope is only as good as the sensors that prove you are holding it, and they go at the IT inlet, not on a wall or in the return. The standard practice is inlet sensors at the rack face, and on a tall or dense rack, more than one, because the top of the rack runs hotter than the bottom. ASHRAE guidance has long suggested reading the inlet at multiple heights for exactly this reason.
Put sensors at the top, middle, and bottom of the representative racks, and concentrate them on the worst spots: the ends of rows, the tops of the densest cabinets, and anywhere the airflow has to turn a corner to arrive. A handful of well-placed inlet sensors that catch the hot spots beats a hall full of sensors that all sit in the easy middle of the cold aisle.
A data center infrastructure management platform, the DCIM, is where those readings turn into something you can act on. It trends the inlets over time, alarms when one drifts toward the limit, and ties the temperature picture to the power and airflow so you can see a hot spot forming before it throttles a server. The platform also holds the record that proves the hall stayed in the envelope, which is what an owner and a commissioning agent ask for. Tie the inlet sensors, the cooling-unit data, and the power meters into one view rather than reading three systems that do not talk.
Top-of-rack hot spots and the uneven inlet
The inlet temperature is not the same across a rack, and the top is almost always the hottest point. Hot exhaust rises, recirculation pulls it back over the top of the cabinet first, and the densest gear often lives high in the rack, so the top inlet is where the envelope gets violated while the room and the bottom of the rack still read fine. Find the worst inlet and you have found the constraint.
The classic hot spot is the top of an end-of-row cabinet in a hall without full containment, or a rack with open U-spaces that let air loop through the cabinet instead of past the servers. Missing blanking panels turn a rack into a short circuit for its own exhaust. The fix for most hot spots is air management, not more cooling: blank the open spaces, seal the floor cutouts, and contain the aisle so the cold air has nowhere to go but through the gear.
Design and commission to the worst inlet, never the average. If the top of the worst rack sits at the edge of the band at full load, the hall is at its limit even if the average says there is room. The whole envelope is a worst-case standard, because the rack that overheats does not care what the average inlet was.
Short excursions into the allowable range
The allowable range is built for excursions, the short periods when the inlet rides above the recommended band, and the gear tolerates them as long as they stay short and inside the class limit. An economizer peak on a hot afternoon, a cooling unit dropping out in an N+1 event, a maintenance window: these are what the allowable band absorbs. They are events, not a steady state.
ASHRAE's framing on excursions ties to time at temperature, the same idea behind the x-factor. A few hours per year near the allowable ceiling cost very little reliability when the rest of the year runs cool. Constant operation near that ceiling is a different bargain, because the failure-rate increase that was trivial for a handful of hours becomes a real number when it runs all year. The allowable band is a budget of time, not an open-ended permission.
The practical takeaway is to design the cooling so that a single failure pushes the inlets into the allowable band for a bounded time, not over the allowable ceiling at all. If a CRAH dropping out sends the worst inlet past the allowable limit, the redundancy was undersized for the setpoint you chose. Prove the failure case before compute arrives, which is exactly what the CRAC/CRAH airflow commissioning is for.
High density, AI, and the shrinking air envelope
High-density and AI workloads are changing the conversation, because the heat from a rack of accelerators is more than air can practically remove. A traditional rack might draw a handful of kilowatts. An AI training rack can draw ten times that or more, and at those densities the air-side envelope alone cannot keep the chips inside their limits. The air does not disappear from the problem, but it stops being the whole problem.
What is emerging is a split. Liquid carries the heat off the hottest components, the processors and accelerators, and air still handles the rest of the rack and the room: the power supplies, the networking, the drives, and the equipment that is not liquid-cooled. So the ASHRAE air envelope still governs the air-cooled portion even in a liquid-cooled hall, and the inlet rule still applies to whatever the air is feeding.
For the air-cooled gear, nothing in this guide changes. The recommended and allowable bands, the inlet measurement, the dew-point control, and the worst-case hot spot all still hold. What changes is that a growing share of the heat leaves through a different path, which is its own envelope with its own targets, covered under liquid cooling rather than here.
How liquid cooling ties in
Liquid-cooled IT runs to its own set of temperature targets, separate from the air envelope, and the headline is that the liquid does not have to be cold. ASHRAE TC 9.9 also publishes liquid-cooling classes with supply-water temperature ranges, and the warmer classes allow water well above room temperature, which is why this is often called warm-water cooling.
The reason warm water works is the same reason warm air works, taken further. Water carries far more heat per unit volume than air, so even tepid water pulls enormous heat off a processor cold plate. Supplying that water warm means the cooling plant can reject the heat with economizers and dry coolers for most or all of the year, often without mechanical chilling at all. The efficiency case for liquid is largely this.
For the purposes of the air envelope, treat the liquid loop as a parallel system with its own setpoints to verify against the equipment and the TC 9.9 liquid classes. The air-cooled remainder of the hall still has to sit inside the air envelope this guide describes.
Why the setpoint pays: economizers and PUE
Power usage effectiveness, the PUE, is the ratio of total facility power to the power the IT gear actually uses, and cooling is the largest piece of the overhead that drives it above 1.0. Raising the supply setpoint and adding economizer hours attacks that overhead directly, which is why the envelope is an efficiency tool and not just a reliability constraint.
The chain is short. A warmer supply temperature lets the chillers work less and run more efficiently, and it widens the window where the outside air or the cooling towers can carry the load with little or no mechanical cooling. More economizer hours mean fewer chiller hours, and the fans can slow when the air does not have to be pushed as hard. Each of those drops the cooling overhead, and the PUE falls toward 1.0.
The well-run fleets that report PUE near 1.1 to 1.2 get there with exactly this combination: contained aisles, warm supply held to the worst inlet, aggressive economizing, and controls that watch the inlets in real time. None of it is exotic. It is the envelope, measured at the inlet and held with containment, turned into an energy result. A PUE deep-dive is its own topic, but the setpoint is where it starts.
Commissioning the envelope before compute arrives
The envelope is a commissioning deliverable, not an assumption, and the time to prove it is before the hall is loaded with revenue compute. The commissioning agent verifies that the inlet temperatures across the hall sit inside the recommended band at design load, that the humidity holds inside the dew-point envelope, and that no rack inlet, including the worst top-of-rack spot, is over the line.
The setpoint gets verified, not just set. Confirm the supply temperature is where the design intended, confirm the cooling units are controlling to the inlet or supply signal rather than chasing the return, and confirm the controls agree across units so they are not fighting each other on temperature or humidity. Then run the failure cases: drop a cooling unit and prove the remaining capacity holds the worst inlet inside the allowable band for the required time.
The artifact that comes out of this is the record an owner keeps: the inlet map at load, the setpoint, the humidity trend, the hot-spot list and what was done about each one, and the failure-mode results. That record is what proves the hall was accepted inside the envelope, and it is what the next engineer reads when a rack runs warm a year later and the question is whether it was ever right.
What to document
An unrecorded thermal envelope leaves you with nothing to point to the day a rack throttles and the setpoint gets blamed. The record is what answers the question months later when a rack throttles and someone asks whether the setpoint was ever correct for the gear in that aisle.
Capture the equipment class for each aisle, the recommended and allowable limits you held to, the supply or cold-aisle setpoint, the worst-case inlet reading at full load and where it was, the humidity band as dew point with the relative-humidity context, the sensor locations, and the failure-case results. Write down which class governed each aisle and why, because the next person will want to know whether the setpoint had room to rise or was already at the limit.
| Parameter | Range or value | Note |
|---|---|---|
| Equipment class (per aisle) | A1, A2, A3, or A4 | Lowest class in the aisle governs |
| Recommended temperature | About 18 to 27 C (64 to 81 F) | Steady-state target at the inlet |
| Allowable temperature | By class, up to the class ceiling | Short excursions only |
| Supply / cold-aisle setpoint | Set to hold worst inlet in band | Design to the worst case, not the average |
| Worst-case inlet at load | Measured, with location | Top of densest or end-of-row rack |
| Humidity (dew point) | Per TC 9.9 envelope | Control to dew point, not RH alone |
| Sensor locations | Inlet, top/middle/bottom | Concentrated on hot spots |
| Failure-case result | Worst inlet during unit loss | Must stay inside allowable for required time |
Common mistakes
- Running the hall too cold out of habit, paying for cooling the equipment never needed.
- Measuring the room average or the return air instead of the IT equipment intake at the rack.
- Holding a tight humidity band so units humidify and dehumidify against each other and waste energy.
- Raising the supply setpoint without containment, so the worst inlet goes over the line on a hot day.
- Ignoring the top-of-rack and end-of-row hot inlet and designing to the average instead of the worst case.
- No inlet sensors and no DCIM trend, so a hot spot is found by a throttled server instead of an alarm.
- Treating the allowable range as a higher steady setpoint instead of a time-limited excursion band.
- Controlling humidity to relative humidity alone, which chases a moving target as supply temperature changes.
Field checklist
Want this checklist to run itself on every job — with photo proof and a signed record crews can hand the customer? That's FieldOS.
Standards and references
ASHRAE Technical Committee 9.9 is the source for the thermal envelope. Its Thermal Guidelines for Data Processing Environments defines the equipment classes A1 through A4 plus the high-density air-cooled class H1, the recommended temperature range of about 18 to 27 C, the wider allowable ranges by class, and the humidity envelope expressed in dew point with a relative-humidity context. The same body publishes the x-factor reliability data and the liquid-cooling classes. The recommended and allowable numbers shift between editions, so confirm them against the current edition before you cite them on a submittal.
The equipment manufacturer is the other authority, and where the manufacturer lists a tolerance tighter than the class, the manufacturer governs. The equipment's rated class and any documented limit control the setpoint for that gear. Cite the class from the nameplate or the data sheet, not from memory of what that line of servers usually is.
For operations and uptime practice, Uptime Institute and the broader data center operations literature cover how to run the hall, verify the redundancy, and hold the envelope under failure. Hedge the specific ranges and classes to ASHRAE TC 9.9 and the equipment manufacturer, and on every job hold two rules above the rest: measure the envelope at the IT inlet, and raise the setpoint only with containment and monitoring in place.
Units, terms, and conversions
The envelope shows up in a few unit systems and a few names, so the same condition can read differently across an ASHRAE table, a manufacturer sheet, and a controls screen.
Temperature is in degrees C in the ASHRAE guidelines and often degrees F on equipment and controls in North America: 18 to 27 C is 64 to 81 F. Moisture is given as dew point in degrees C or F and as relative humidity in percent, and the two are not interchangeable, because relative humidity depends on temperature while dew point does not. Read the intake condition, sometimes written inlet or supply air temperature, at the front of the rack. The class, A1 through A4, is a property of the equipment that sets how wide its allowable range is.
- Thermal envelope
- The temperature and humidity range allowed at the IT equipment intake, per ASHRAE TC 9.9
- Recommended range
- The conservative steady-state band, about 18 to 27 C (64 to 81 F) for all classes
- Allowable range
- The wider band the equipment tolerates, set by class, intended for short excursions
- Equipment class (A1-A4)
- ASHRAE classification by allowable range width; A1 tightest, A4 widest
- Dew point
- The temperature at which air saturates; a measure of actual water content, independent of air temperature
- X-factor
- ASHRAE's relative hardware failure rate indexed to a 20 C inlet baseline of 1.0
- Inlet / intake air
- The air entering the front of the IT equipment in the cold aisle, where the envelope is measured
- PUE
- Power usage effectiveness, total facility power divided by IT power; lower is more efficient
FAQ
What is the ASHRAE recommended temperature for a data center?
The ASHRAE TC 9.9 recommended temperature range is about 18 to 27 C, or 64 to 81 F, measured at the IT equipment intake in the cold aisle. It is the same for classes A1 through A4 and is the conservative steady-state target. The equipment manufacturer's listed limit controls where there is a tighter one.
What is the difference between recommended and allowable?
The recommended range, about 18 to 27 C, is the conservative band for steady operation. The allowable range is wider and set by equipment class, meant for short excursions like a hot afternoon or a cooling unit out for service. Hold the inlets in the recommended band and let them ride into allowable only briefly.
Can you raise the data center temperature to save energy?
Yes, and many halls run too cold. Raising the supply temperature toward the top of the recommended band, near 27 C at the inlet, cuts cooling energy and adds economizer hours. Do it with hot-aisle or cold-aisle containment and inlet monitoring in place, raising it in steps while watching the worst rack inlet for trouble.
Where do you measure data center temperature?
Measure it at the IT equipment intake, the front of the rack in the cold aisle, not the room average or the return air. The server only experiences its inlet air. On a tall rack, read top, middle, and bottom, because the top usually runs hottest. Design and accept the hall on the worst inlet, not the average.
What humidity should a data center run at?
ASHRAE TC 9.9 sets the recommended moisture band in dew point with a relative-humidity context, commonly cited up to about 60 percent relative humidity with a dew-point ceiling near 15 C and a minimum dew-point floor. Too dry risks electrostatic discharge; too wet risks condensation and corrosion. Confirm the current edition numbers against the equipment.
Why control humidity by dew point instead of relative humidity?
Dew point measures actual water content, which does not change as air warms moving through the hall, while relative humidity does. Controlling to relative humidity makes units chase a moving target and add moisture the gear never needed. Dew-point control with a wide band cuts humidification and dehumidification energy and stops units fighting each other.
Does running a data center warmer hurt reliability?
Modestly, and ASHRAE quantifies it with the x-factor, a failure rate indexed to a 20 C baseline of 1.0. Failure rate rises with temperature but stays low in absolute terms. With an economizer, cool hours offset hot hours, so the net annual reliability of a warm hall can match a constantly cool one. Weight your own climate.
Do you need containment to raise the setpoint?
Effectively yes. Without containment, hot exhaust mixes back into the cold aisle, so the worst inlet runs several degrees above the average you measure. Raise the setpoint blind and that hidden rack goes over the line first. Containment makes the inlet uniform so the headroom you measure is real, which is why it comes before any setpoint increase.
What is an ASHRAE equipment class A1 to A4?
ASHRAE TC 9.9 classes A1 through A4 sort IT equipment by how wide an allowable temperature range it tolerates, with A1 the tightest and A4 the widest. Most mainstream servers are A2, allowing intake up to roughly 35 C. The class is set by the manufacturer, and in a mixed aisle the lowest class present governs the limit.
How do AI and liquid cooling change the air envelope?
High-density AI racks make more heat than air can practically remove, so liquid carries heat off the processors while air still cools the rest of the rack and room. The ASHRAE air envelope still governs the air-cooled gear and the inlet rule still applies. Liquid loops run to their own warmer water-temperature targets, verified separately.
People also ask
Codes cited in this guide
This guide is written and reviewed against the published standards below. Always confirm the current adopted edition with the authority having jurisdiction.