Rear-door heat exchanger commissioning field guide for high-density racks

Rear-door heat exchangers, and where they fit among the liquid options

A rear-door heat exchanger is a water coil that takes the place of a rack's rear door and pulls the heat out of the server exhaust at the rack, before that hot air ever mixes into the room. The hot air leaving the servers passes through the coil, the water in the coil carries the heat off to a coolant distribution unit or the facility chilled water, and the air that comes out the back is close to room temperature. The rack cools itself instead of dumping its heat into the hall for a CRAH to chase down later.

Where it fits among the liquid options is the part people get wrong. A rear-door exchanger is the lightest-touch way to push a rack past what air alone can carry, because it bolts onto an existing rack and never touches the server internals. Direct-to-chip runs coolant onto a cold plate clamped to the silicon and carries far higher density, and the direct-to-chip loop commissioning is its own scope. Immersion drops the whole server in fluid. The rear door sits at the entry end of that ladder. It is still cooling air, so it tops out lower than cold plate, but it is the one you can retrofit into an air-cooled hall without re-plumbing a single server.

What this guide commissions is the door and its water connection, not the chip. The cooling pillar covers the air-side architecture and the chilled water plant behind the door, and the direct-to-chip guide covers the cold-plate loop. This one stays on the coil hanging off the back of the rack and the water that feeds it.

Passive or active: which rear-door heat exchanger?

The split is whether the door has its own fans. A passive rear door has no fans and no moving parts. It relies entirely on the server fans inside the rack to push the exhaust air through the coil, which means the coil's air-side resistance lands on those server fans as extra back pressure. A passive door suits moderate density, commonly cited in the range of 20 to 30 kW per rack where the airflow path is clean and the server fans have the static pressure to spare. Confirm the number against the door manufacturer, because the passive ceiling moves with the coil and the servers in front of it.

An active rear door adds its own fans, usually EC fans, that draw the exhaust through the coil and take the coil's pressure drop off the server fans. That is what lets the active door carry higher density, commonly into the 30 to 50 kW per rack range and beyond on the larger units, with some active doors rated up to roughly 75 kW per rack. The active door also holds its performance more steadily across a mix of server loads, because the door fans, not the servers, set the air-side flow.

The fan interaction is the part that bites in commissioning. A passive door that adds too much back pressure makes the server fans work harder, spin up, and burn fan energy that erodes the efficiency the door was supposed to buy. An active door whose fans are tuned wrong can fight the servers: run the door fans too slow and they choke the rack, run them too fast and they pull air backward through a lightly loaded server and disturb its own thermal control. The active door's fan speed has to be tuned to track the rack, which is a commissioning task, not a factory setting. Verify the passive-versus-active question and the fan dependence on the actual equipment, since the two architectures fail in opposite directions.

The water side: supply, return, and flow per door

The water side is what the door rejects heat into, and it comes from one of two places: a coolant distribution unit that isolates a clean loop from facility water, or the facility chilled water directly. A CDU buys the same isolation it buys on a direct-to-chip job, a clean conditioned loop at the door and the dirtier building water kept on the other side of a heat exchanger. Feeding the door straight off facility water is simpler and shows up on smaller deployments, at the cost of putting building water quality right at the coil.

The numbers that govern are the supply and return temperature, the flow rate per door, and the resulting capacity. Each door takes a design flow, given in gallons or liters per minute, set to carry the rack's kilowatts at the design temperature rise across the coil. Too little flow and the coil cannot pull the rack's heat, the exhaust comes out warm, and the rack's neighbors inherit it. The supply temperature is the lever that decides almost everything else about the door, and it is set deliberately warm, for the reason the next section covers.

The one number that is not negotiable is the relationship between the supply water temperature and the room dew point. Run the coil water below the room dew point and you condense water out of the air onto the coil. Everything about how the water side is set up, the supply temperature, the control valve, whether you need a condensate provision, traces back to that one constraint.

Why run the coil water above the room dew point?

You run the coil water above the room dew point because a coil colder than the dew point condenses water out of the air, and a wet coil at the back of a powered rack is the failure you are commissioning against. When the supply water drops below the dew point, moisture in the exhaust air condenses on the coil fins the same way it beads on a cold drink, and now you have liquid water forming on a structure that swings on hinges directly behind energized electronics. That is the central risk of a rear-door heat exchanger, more than the leak in the pipe.

The control answer is to hold the chilled water supply above the room dew point at all times. Common practice keeps the supply water above the data hall dew point, often in the range of roughly 18 to 20 C depending on the room conditions, with a modulating control valve that watches the entering water temperature and the dew point and never lets the supply slip below it. The dew point, not the relative humidity, is the number that governs condensation, which is the same reason modern halls control to dew point on the air side.

If the design does run the coil below dew point, for a high-capacity active door chasing more heat than warm water can carry, then it needs a condensate provision: a drip pan under the coil, a drain or a condensate pump, and the piping to carry the water away. Most rear-door designs are set up specifically to avoid that, because a condensate drain at the back of a rack is one more thing to clog, overflow, and drip onto hardware. Running above dew point eliminates the condensate problem instead of managing it, and that is the design most owners want. Verify the dew point margin and the condensation control against the room's actual dew point, because it is the single most consequential setpoint on the door.

Warm-water, sensible-only operation and free cooling

A rear-door heat exchanger does sensible cooling, which is to say it lowers the temperature of the exhaust air without pulling moisture out of it, by design. Holding the coil above dew point means no condensation, which means the door changes the air's temperature and not its moisture content. The room's humidity stays the job of the CRAC or CRAH units that handle the residual load and the latent control, while the door takes the sensible heat the rack throws off.

Running warm is where the energy case lives. Because the coil only has to be cooler than the hot exhaust air, not cold in absolute terms, the supply water can sit in the high teens to high 20s of degrees C and still pull the rack's heat, since heat flows on the difference between the exhaust and the water and that difference stays large. Warmer chilled water means the plant can make it with a cooling tower or a dry cooler for far more hours of the year, and in many climates skip mechanical refrigeration for the door load. The economizer and the chilled water plant that make this pay off are covered in the cooling pillar. The point at the door is that cold water buys nothing the warm water does not, and it costs chiller energy and risks condensation on the way.

This is the same logic that runs a direct-to-chip loop warm, one architecture over. The rear door just applies it to air instead of a cold plate. Set the supply as warm as the rack's heat and the coil's capacity allow, hold it above dew point, and let the economizer carry the hours.

Piping and the swinging-door connection

The piping to a rear door has one problem a fixed coil does not: the door swings. The supply and return have to reach a coil mounted on a hinged door that opens 90 degrees or more for service, so the connection has to flex through that arc thousands of times without leaking. The usual answer is a flexible hose loop or an articulated connection at the hinge, sized and routed so the door opens fully without kinking the hose or straining the fitting.

The connection itself is built for service. Quick-disconnects, often dripless couplings, let a door be isolated and pulled without draining the loop or shutting the row, the same parts and the same logic as the quick-disconnects on a direct-to-chip manifold. Isolation valves on the supply and return let one door come off line while the rest of the row keeps running. A manifold in the row or in the cabinet feeds the doors off the branch.

The leak risk concentrates exactly where the motion is. The flexible hose at the hinge, the quick-disconnects, and the threaded or pushed fittings at the door are where an aging rear-door loop starts to weep, not in the middle of a straight pipe run. That is the same lesson the direct-to-chip loop teaches: leaks start at the connections and the couplings that move, which is why the leak detection and the drip containment get aimed there first. Check the flexible connection's clearance through the full door swing at install, because a hose that just clears at 80 degrees gets pinched at 95 and fails six months later.

Pressure test and flush before connecting

The coil and the loop get proof-tested and flushed before they carry coolant near hardware, the same hold points as any chilled water or liquid cooling loop. The proof test pressurizes the assembled coil, the hoses, and the connection above the system working pressure, holds it for a documented duration, and watches the gauge for any decay that betrays a leak. The test pressure, the medium, and the hold time come from the project specification and the applicable piping code, commonly an ASME B31 section by topic, not from habit. Hold the test long enough for the test fluid temperature to stabilize, because a falling gauge can be a real leak or just the water cooling and contracting, and the two look the same for the first few minutes.

Flush before you connect. New coils, hoses, and manifolds come full of the debris any jobsite piping carries, scale, cutting oil, thread sealant, and grit, and the rear-door coil has narrow fin passages and a control valve that fouling will choke. Circulate the loop through filtration until it meets the cleanliness the spec calls for, then connect. The discipline is the same the direct-to-chip guide treats as the make-or-break step and the cooling pillar treats as a hold point on the chilled water side. The consequence is smaller here than on a cold plate, because a rear-door coil is more forgiving than a micro-channel, but a fouled coil or a stuck valve still costs you capacity quietly. The HydroTest workflow walks through holding the pressure, logging the decay, and telling a true leak from a thermal settle.

How do you leak-test a rear-door heat exchanger?

You leak-test a rear-door heat exchanger in two steps: prove the loop holds pressure, then prove the detection catches a leak that does happen. The pressure test above finds the leaks you can prevent. The leak-detection test proves the system that protects the hardware actually works, because a rear door hangs a water-filled coil and hoses directly behind a powered rack, and a leak here lands on energized electronics.

Detection layers the same way it does on any liquid loop. A leak sensor, rope cable along the piping or a spot sensor at the low point, reports a leak and roughly where it is. A drip tray or containment under the coil and the wet connections gives a leak somewhere to go that is not the equipment below. Commissioning means actually wetting a sensor with a measured drop and confirming it alarms, the alarm reaches the BMS, and where the design calls for it the valve isolates the door. A detection system nobody exercised by wetting a sensor is a system you are only hoping works.

Some designs run the door loop at negative pressure, below ambient, so a breach draws air in rather than pushing water out onto the rack, the same inside-out trick the direct-to-chip guide covers. If that is the architecture in front of you, the test changes: you prove the loop holds vacuum and that an induced leak shows up as the air-ingress signature, not as water on a tray. Know which one you have before you set the test, because the failure mode you are proving against is the opposite in each.

Airflow, neutral pressure, and the thermally neutral room

The point of a rear door, done right, is a thermally neutral rack: the coil pulls close to 100 percent of the rack's heat out of the exhaust, so the air leaving the back of the rack comes out near room temperature and the rack adds almost no net heat to the hall. A row of well-tuned rear doors can make a room thermally neutral, which is why a rear-door deployment often needs no containment, no ceiling plenum, and far less room-level air handling than the same load on raised-floor air. The heat goes to water at the rack instead of into the room.

Getting there is an airflow-balancing job, and the active door is where it gets delicate. The door fans have to move close to the same air the servers move, no more and no less. Tune them to pull too hard and the active door starves the servers behind it or pulls room air backward through a lightly loaded server and confuses its onboard fan control. Tune them too soft and the exhaust backs up, the coil cannot capture it all, and warm air spills into the room. The target is a near-neutral pressure across the door, where the door neither starves nor over-pulls the rack, tracked across the rack's load range, not at one operating point.

A passive door does not have this lever, which is both its simplicity and its limit. It moves exactly the air the server fans push, so it cannot starve the servers, but it also cannot help them, and the coil's back pressure makes those server fans work harder for the air they do move. Verify the near-neutral, full-heat-capture behavior under load, because a door that looks balanced at idle can spill heat once the rack ramps.

How much heat can a rear-door heat exchanger reject?

A rear-door heat exchanger rejects on the order of 20 to 30 kW per rack passive and 30 to 50 kW and up active, with the largest active doors rated to roughly 75 kW per rack, but the door manufacturer's curve governs and the real number moves with the water temperature and flow you give it. Those ranges are where the products cluster as of this 2026 review, not a law, so confirm the rated capacity of the actual door against its performance data, hedged to the water temperature and flow at your site.

The capacity is not a single number, it is a curve. A door rated at 50 kW is rated at a stated supply water temperature, flow rate, and entering air temperature. Give it warmer water to chase free-cooling hours, or less flow than design, and the capacity comes down. This is the trade at the heart of warm-water operation: the warmer you run the supply for energy, the less heat the coil pulls per degree, so the design picks a supply temperature that holds capacity while staying above dew point. Match the door's rated capacity at the site's actual water conditions to the rack's real load, not the door's headline number to the rack's nameplate.

Part-load is the easy direction. A door sized for a full rack handles a half-loaded rack with room to spare, and the control valve simply throttles the water back. The hard direction is a rack that grows past the door, which on a rear-door deployment usually means the rack has outgrown air-side cooling entirely and wants direct-to-chip, not a bigger door.

Controls: the valve, the fan speed, and the failure mode

The control on a rear door is a water valve that modulates to hold a target, and the target is usually the air temperature leaving the coil or the rack inlet on the next row. As the rack load rises, the valve opens to put more water through the coil and pull more heat; as it falls, the valve throttles back. The dew point limit overrides all of it: the valve and the supply temperature are held so the coil water never drops below the room dew point, no matter what the temperature loop is asking for.

On an active door the fan speed is the second control, usually tracking the rack airflow or a differential pressure across the door so the fans move the servers' air without fighting them. The two loops, water valve for heat and fan speed for airflow, have to be tuned together, because changing one moves the other's operating point. The BMS integration ties it up: the exhaust temperature, the water temperature, the valve position, the fan speed, and the leak alarm all have to reach the BMS and read correctly, and the critical alarms have to drive the design response. An alarm that lives only on the door's local controller is an alarm nobody sees at 3 a.m.

The failure mode is the one to design and test for. When a door fails, loses water, loses fan power, or has its valve stick shut, it stops capturing heat and the rack's full exhaust dumps straight into the room. On a hall built around thermally neutral rear doors with little room cooling, that heat has nowhere to go and the surrounding racks heat up fast. So the design needs a fail-over: backup room cooling that can carry the load through a door failure, or enough redundancy in the doors and the water plant that one door dropping does not strand a rack. Prove the failure response, do not assume it, because a thermally neutral room has the least margin exactly when a door quits.

The commissioning sequence

Rear-door commissioning runs the same build-up as any cooling system: static checks first, then function, then load and failure. The order matters because each step proves the ground the next one stands on, and skipping a hold point means finding the problem later with hardware in front of it.

Mechanical install and pressure test come first. The door is hung, the flexible connection clears the full swing, and the coil and loop are proof-tested to the spec and the applicable piping code before anything else. Flush the loop to the cleanliness target, then fill and vent, because air trapped in a coil is capacity you paid for and are not getting, and a spot where corrosion starts. Balance the flow to each door to its design rate, the same per-door discipline the direct-to-chip guide applies per node, so no door in the row is starved while another runs rich.

Then the controls go functional: the water valve modulates to the air target, the dew point limit holds the supply above the room dew point, and on an active door the fan speed tracks the rack. Test the leak detection by wetting a sensor and confirming the alarm and the isolation response. Then load it, with IT load or simulated heat, and verify the door captures the rack's heat and holds the room near thermally neutral at full load, not just at idle. Last, prove the fail-over: drop a door and confirm the backup cooling or the redundancy carries the rack. Verify the sequence against the manufacturer's commissioning procedure and the project spec, since the exact steps and hold points move with the equipment.

Maintenance the owner inherits

A rear door is not fill-and-forget, and the turnover package should say so. The coil collects dust on the air side the same as any finned coil, and a fouled coil loses capacity quietly until a rack runs warm and nobody knows why. Coil cleaning on a schedule keeps the rated capacity available. The water side carries the same chemistry burden as any chilled water or treated loop: the water treatment, the corrosion inhibitor, and the biological control deplete over time, and a neglected loop fouls the coil and the valve from the inside. Water treatment is its own topic and the cooling-tower and water-treatment work covers it by subject; the point here is that the door inherits it.

On an active door the fans and any filter are the wear items. EC fans run for years but they do fail, and a door that quietly loses a fan loses the capacity that fan was moving. Where the door has a filter, it loads up and needs changing on a schedule, or it adds back pressure the design did not plan for. The leak checks never stop: the flexible hose at the hinge and the quick-disconnects are the parts that move and weep, so they get looked at, not assumed.

Hand the door over with a maintenance plan, a coil-cleaning interval, a water-treatment and sampling schedule, a fan and filter interval, and a leak-check routine, the same way the direct-to-chip loop is handed over with a chemistry baseline. A door turned over without that plan degrades until something runs warm, and by then the cause is buried.

RDHx vs direct-to-chip vs CRAH: when does each fit?

The three sit on a ladder of density and how invasive the change is. A CRAH, the room air handler over a chilled water coil, cools the whole hall through the air and tops out where air-side density tops out, roughly the tens of kilowatts per rack before mixing and airflow defeat it. A rear-door heat exchanger bolts onto an existing rack and pushes it higher, commonly into the 20 to 50 kW range and up, without touching the servers, which makes it the most retrofit-friendly liquid step there is. Direct-to-chip runs coolant onto a cold plate on the silicon and carries the highest density of the three, well past what air can do, at the cost of re-plumbing the server and a much more demanding clean-and-leak regime.

The retrofit case is what sells the rear door. You can take an air-cooled hall, hang doors on the hot racks, run a water loop down the row, and push density up without opening a single server or rebuilding the room's containment, since a thermally neutral door needs no containment. That is a weekend per rack, not a server redesign. The ceiling is that you are still cooling air, so the rear door runs out of room before the cold plate does.

The honest framing is that these layer, they do not compete. Many AI deployments run direct-to-chip on the GPUs and a rear door or contained air on the rest of the rack heat the cold plate does not touch, which is why the air-side and rear-door commissioning does not go away when cold plate arrives. Pick the rear door when the density is past comfortable air but you want to keep the servers and the room as they are. Pick the cold plate when the rack has outgrown air entirely. The direct-to-chip and the cooling-pillar guides cover the other two rungs.

The data center deployment: rows, facility water, and redundancy

At the scale of a hall, a rear-door deployment is a row of doored racks fed off a water loop, and the design questions move from the single door to the row and the plant behind it. Each row of doors draws a total flow and rejects a total load that the facility water or the CDUs serving the row have to carry, and that load is real cooling capacity that has to exist back at the chillers or the dry coolers, not just at the door. A row of forty doors at 40 kW is 1.6 MW of cooling the plant has to make and reject. The rear door only moved where the heat changes from air to water; it did not make the heat smaller.

Facility water capacity is the constraint that decides how far the deployment scales. The doors are cheap and fast next to the chilled water plant, the pumps, and the heat rejection behind them, so a retrofit that adds doors faster than the plant can serve them runs out of water before it runs out of racks. Size the plant and the distribution for the row total at the design water temperature, and remember the warm-water operation that helps the energy case also means the plant rejects at a warmer, easier condition.

Redundancy follows the same N, N+1, and 2N logic as the rest of the cooling chain, applied to the doors, the pumps, the CDUs, and the water path. The failure mode makes it sharper here: because a failed door dumps its rack's heat into a room with little spare cooling, the deployment needs either door-level and plant-level redundancy or a backup room-cooling layer that catches the load when a door drops. A row of thermally neutral doors with no fail-over is a row that overheats together the moment the water plant hiccups. The integrated test that proves the plant and the power ride through a failure together is the cooling-pillar and power-QA scope, and the rear-door row is one more load riding on it.

Field example: a passive door that starved on its own back pressure

A retrofit hung passive rear doors on a row of racks that had crept from 12 kW to about 28 kW as the hardware refreshed, and the row started throwing high server-fan-speed alarms and intermittent high-inlet alarms on the racks downstream. The first instinct was that the doors were undersized for the new load. The water numbers said the doors had capacity. The air numbers told the real story.

The passive coil's back pressure at 28 kW was more than the server fans had been built to push through, so the fans ran near full speed to move their air through the coil, and at the top of two racks the exhaust was not fully captured and spilled warm air that the next row pulled in. Meanwhile the chilled water had been set cold, around 12 C, to chase capacity, which put the coil below the room's roughly 14 C dew point and left the bottom of two coils damp at the morning humidity peak. The door was solving the heat by creating two new problems: starved server fans and a coil running under dew point.

The fix was to convert the two highest racks to active doors and raise the supply water to about 18 C, above the dew point, with the valve modulating to the air target. The active fans took the coil back pressure off the server fans, the server fans dropped back to normal, the warm spill and the downstream alarms cleared, and the warmer water ended the condensation. Same racks, same loop, two doors changed and one setpoint corrected. The lesson the row taught is that a rear door fails on the air side and the dew point long before it fails on water capacity.

What to document

A rear door that was proven but never documented hands operations a row of black boxes and turns the first warm rack into a guessing game. The record is what tells the next engineer whether a reading is a new problem or how the door has always run. Capture, per door, the door identity and type, the water supply and return temperatures, the flow, the dew point margin the supply is held above, the leak-detection test result, and the verified capacity at the site's water conditions.

Two records carry the most weight later. The dew point margin proves the coil was set to run dry, which is the question behind every condensation worry an operator will ever raise, and the leak-detection test proves the system protecting the hardware was actually exercised. A turnover package missing either leaves the owner trusting that the two most consequential steps got done.

Common mistakes

• Running the coil water below the room dew point and condensing water onto a coil that swings behind a powered rack.
• Setting the chilled water cold to chase capacity instead of running it warm above dew point with the valve modulating.
• Connecting the door without flushing the loop or proof-testing the coil, so the first leak shows up over live hardware.
• Tuning an active door's fans too hard and starving the servers, or too soft and spilling warm exhaust into the room.
• Hanging a passive door at a density past what the server fans can push through the coil's back pressure.
• Installing leak detection but never wetting a sensor to prove the alarm and the isolation response fire.
• Building a thermally neutral room with no fail-over, so a single failed door overheats the racks around it.
• Letting the coil foul or the water treatment lapse, so the door loses capacity quietly until a rack runs warm.
• Balancing to the row average instead of the per-door flow, leaving one door in the row starved.

Field checklist

Standards and references

The thermal framework comes from ASHRAE Technical Committee 9.9, whose thermal guidelines set the data hall temperature and humidity envelope, the dew point the coil has to stay above, and the liquid-cooling water classes that frame warm-water operation. The dew point control that keeps the coil dry is an application of the same TC 9.9 humidity and dew point guidance the air side runs on. The Open Compute Project publishes rack and door requirements many deployments design to, including a door heat exchanger requirement for open rack by topic, so name the OCP document by subject and confirm the current revision, since it revises on its own cycle.

The capacity, the flow, the design water temperature, and the dew point setpoint are governed first by the door manufacturer's specification and performance data, which override any general guideline where they are stricter. The pressure test of the piping falls under the applicable pressure-piping code, commonly an ASME B31 section by topic, together with the project specification that sets the test pressure, medium, and hold. The commissioning process framework draws on the ASHRAE commissioning guidance, commonly Guideline 0, and where a facility is chasing a tier, the Uptime Institute standards drive the witnessed redundancy and integrated-test demonstrations.

Edition numbers, the exact water-class values, and the OCP revision levels move between cycles, so confirm them against the published documents and the door manufacturer before citing them on a submittal. ASHRAE TC 9.9 and OCP give the framework; the door manufacturer's capacity and water specifications and the project documents control the limits you commission to.

Units, terms, and acronyms

Rear-door cooling borrows vocabulary from HVAC, from the chilled water plant, and from the IT side, and the same idea reads differently across a door submittal, a CDU datasheet, and an ASHRAE guideline. The terms below travel across the rear-door scope.

RDHx

Rear-door heat exchanger, a water coil replacing a rack's rear door that captures the server exhaust heat at the rack

Passive RDHx

A rear door with no fans that relies on the server fans to push exhaust air through the coil

Active RDHx

A rear door with its own fans that draw exhaust through the coil, taking the back pressure off the server fans

Dew point

The temperature at which moisture condenses out of the air; the coil supply water is held above it to keep the coil dry

Sensible cooling

Lowering air temperature without removing moisture; what a rear door does when it runs above dew point

CDU

Coolant distribution unit, the pumps, controls, and heat exchanger that can isolate a clean door loop from facility water

Quick-disconnect (QD)

A dripless coupling that lets a door be isolated and serviced without draining the loop

kW per door

The heat a single rear door rejects, set by the water temperature, flow, and the door's capacity curve

Thermally neutral

A rack whose rear door captures nearly all its heat, so its exhaust leaves near room temperature

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What is direct-to-chip liquid cooling?
• What is bypass airflow in a data center?
• What is a CDU in a data center?
• Why do data centers need liquid-cooling leak detection?
• What is data center waste heat reuse?