In-row and close-coupled cooling commissioning field guide for data centers

In-row cooling, and the short air path that defines it

In-row cooling is a cooling unit set in the row between the cabinets, instead of around the perimeter of the room, so it sits inches from the racks it serves. The unit pulls hot exhaust straight out of the hot aisle, cools it across a coil, and pushes cold air back into the cold aisle right where the server inlets draw it. Close-coupled is the same idea named for what it does: the cooling is coupled close to the heat, not parked at the wall.

The whole point is the length of the air path. A perimeter unit blows cold air across the room or under the floor and hopes it reaches the rack inlet before it mixes with hot exhaust. An in-row unit moves the air a few feet, from the hot aisle next door to the cold aisle on its other side, and the shorter that trip, the less the two air streams mix and the less fan energy it takes to move the same heat. The cooling pillar covers the room-level air path and the ASHRAE envelope in full; this guide stays on the unit in the row and what it takes to accept one.

The number that still matters is the temperature at the server inlet, not the air leaving the unit and not the room average. In-row cooling shortens the distance between the supply and the inlet so the two read close to each other, which is exactly the problem perimeter cooling struggles with on a dense floor. Get the row right and the inlet matches the supply. Get it wrong and you have a unit blowing cold air into a row that is still mixing its own exhaust.

Why close-coupled cooling handles more density

Close-coupled cooling carries more kilowatts per rack because the air does less wandering on the way to the heat and back. Every foot of room the cold air crosses is a foot where it can mix with hot exhaust, lose a few degrees, and arrive at the inlet warmer than it left the unit. Shorten that path to the gap between two aisles and the supply reaches the rack almost as cold as it left the coil, so the same unit holds a denser rack inside the envelope.

Fan energy is the other half. Moving air across a room or up through a raised floor and back costs fan power that climbs steeply with distance and resistance. An in-row unit moves air a short, direct route, so for the same heat it spends far less on fans. Field and vendor figures put the fan-energy savings of a close-coupled layout well above what perimeter units spend on a dense floor, sometimes cutting fan power by half or more, though the real number depends on the layout, the containment, and the load.

There is a limit, and it is honest to name it. Air is air. Close-coupling buys you density that perimeter cooling cannot reach, but the rack still has to take in enough air volume to carry its heat, and past a point no amount of close-coupling moves enough air through a single cabinet. That ceiling is where liquid takes over, which the liquid cooling guide covers. In-row is the last and densest air step, not a way to make air do a liquid job.

In-row vs perimeter cooling: which fits the density?

In-row cooling fits the medium-to-high-density row that perimeter cooling cannot keep cold, and perimeter cooling still fits the low-density hall where the cheaper, simpler architecture carries the load. The split tracks rack density more than anything else. Perimeter CRAC and CRAH units are commonly applied below roughly 5 to 7 kW per rack, where the room has time and room to mix the air and still hit the inlet. In-row commonly takes over in the range above that, frequently cited from about 5 to 20 kW per rack, and reaches higher when it is paired with tight containment.

Those numbers are vendor-and-design figures, not a code line, so treat the crossover as a band rather than a hard wall. The binding constraint is whether the room-level air path can deliver cold air to the inlet faster than the racks mix it with their own exhaust. Once a dense row pulls air faster than a perimeter unit can push it cleanly across the room, the supply mixes before it arrives, the inlet runs hot, and the answer is to move the cooling into the row.

Rear-door heat exchangers and overhead units are the neighbors on the same shelf. A rear-door hangs a coil on the back of the cabinet and cools the exhaust before it enters the room, a way to push a rack higher without re-architecting the floor, which the cooling pillar covers alongside the liquid options. Overhead in-aisle units hang the cooling above the aisle to save floor space. They all share the close-coupled logic: get the cooling near the heat. In-row is the version that lives in the row itself and is the most common close-coupled choice on a mixed-density floor.

Does in-row cooling need containment?

In-row cooling needs containment in nearly every real deployment, because the unit only works if it draws genuinely hot air. An in-row unit is built to pull the hottest return it can get and reject that heat efficiently across its coil. Without a sealed hot aisle or a sealed cold aisle, the unit draws a blend of hot exhaust and stray cold supply, the return temperature sags, the coil loses efficiency, and the density the unit was bought for never shows up. Close-coupling plus containment is where the efficiency actually lives.

Hot-aisle containment is the common pairing. You seal the hot aisle so the racks dump exhaust into a closed space, the in-row units along the row draw straight from that hot space, and the cold side of the unit feeds the open cold aisle or the room held at supply condition. The unit gets the warmest possible return, which is exactly what makes the coil and the delta-T work. Cold-aisle containment is also used, sealing the cold supply aisle instead. Either way, the barrier is what stops the supply from short-circuiting back to the return before it has passed through a server. The containment guide covers the seal, the blanking panels, the differential pressure, and the fire-code interaction in full.

Run an in-row unit with no containment and you have spent close-coupled money on a perimeter result. The supply leaks around the racks, the return comes back lukewarm, and the controls chase a temperature the racks never actually see. This is the single most common in-row mistake on the floor: the units go in, the containment gets value-engineered out or never finished, and the row underperforms while everyone blames the units. The containment is not an accessory to in-row cooling. It is the thing that makes the short air path mean anything.

The cooling medium: chilled water, DX, and CDU-fed coils

An in-row unit cools the air across a coil, and what runs through that coil is the first thing to pin down on the submittal, because it sets the whole commissioning scope. Three media show up. Chilled water in-row units take water from a central plant through a coil and modulate a valve to hold the supply or return temperature. Direct-expansion in-row units carry their own refrigerant circuit and a compressor, rejecting heat to a condenser outside, which suits an edge site or a small hall with no chilled water plant. Glycol and pumped-refrigerant variants exist in between.

The chilled water in-row is the workhorse of larger halls, for the same reason CRAH beats CRAC at scale: one efficient central plant feeding many coils beats a field of individual compressors on energy and maintenance. The DX in-row earns its place where running pipe for a chilled water loop is not worth it, or where a self-contained unit is the cleaner install. Read the cut sheet and know which you have, because a chilled water unit puts piping, flow balancing, and leak detection in your scope, and a DX unit puts refrigerant charge, the condenser, and compressor staging in it instead.

Every in-row unit, whatever the medium, shares the same parts worth knowing: a coil that does the heat transfer, fans that move the air, usually electronically commutated EC fans that modulate speed efficiently, and a condensate path for the water the coil wrings out of the air. The EC fans are the reason in-row can turn down so far and save fan energy at part load. The condensate path is the part that gets ignored until it floods a cabinet. Both are commissioning items, not afterthoughts.

The chilled water side of an in-row unit

On a chilled water in-row unit, the loop is the same physics as any hydronic coil, brought right into the row alongside the electronics, which is what raises the stakes. Cold water enters the coil, picks up heat from the return air, and leaves warmer, and a control valve throttles the flow to hold the temperature the unit is set to. The supply and return temperatures, the flow, and the valve action are all things you verify, not assume, because a coil that never gets its design flow cannot make its design capacity no matter how cold the water is.

Leak detection moves up the priority list the moment water runs in the row instead of around it. A perimeter unit that weeps drips on the floor at the wall. An in-row unit that weeps drips next to powered cabinets, so drip trays, point sensors at the low points, and rope sensing along the piping under the row are commissioning items you test by actually wetting a sensor, not by confirming it was installed. The leak detection logic and its alarm path to the building management system are the same discipline the liquid cooling guide details for the technology loop, scaled to a chilled water in-row coil.

Then there is condensate, which is the chilled water side's quiet hazard. A coil running below the room dew point pulls water out of the air, and that water has to go somewhere. The unit either runs sensible-only, holding the coil above the dew point so it never condenses, or it manages condensate with a drain pan, a condensate pump, and an overflow sensor. Confirm the piping is hydrostatically tested and flushed before it feeds a coil, the same hold point the cooling pillar treats as non-negotiable, because the leak you find at commissioning is cheap and the one you find over live gear is not.

Matching the unit cooling to the row load

An in-row layout works when the cooling installed in the row matches the heat the racks in that row produce, and it fails quietly when the two drift apart. The math is the same relationship the air side always lives by: the heat removed equals the airflow times the temperature rise across it times the air's heat capacity, which works out to roughly 160 CFM of air per kW at a 20 F rise. A row pulling 200 kW needs cooling units that can move that heat at that delta-T, sitting in or near that row, not averaged across the whole hall.

The trap is the average. A hall can have exactly enough total in-row capacity on paper and still cook one row, because the heat is not spread evenly and neither is the cooling. The dense AI row at one end pulls far more than the half-empty row at the other, and an in-row design that placed units evenly down both starves the hot row and floods the cold one. In-row cooling is a per-row, per-zone job. You match the units to the load in their zone and you confirm it, row by row, under load.

The other half is that the racks are fans too. They draw a fixed volume through themselves, and if the in-row units in the zone deliver less than that draw, the racks make up the difference by pulling air from wherever they can, which means hot exhaust around the ends or through any gap in the containment. Match the unit airflow to the rack draw with a slight surplus, the same way the containment guide sets a slightly positive cold aisle, and the row holds. Fall short and the row goes negative and starts eating its own exhaust.

How do the in-row units avoid fighting each other?

In-row units avoid fighting each other when they are controlled as a group on a shared signal, not as a row of independent thermostats each chasing its own reading. This is the control problem unique to close-coupled cooling, and it is the one that gets discovered after turnover instead of during commissioning. Put six units in a row, let each one modulate its fan and valve to its own local sensor, and they will hunt against each other: one ramps up, cools its neighbor's sensor, the neighbor ramps down, the first one overshoots, and the row oscillates while burning energy and never settling.

The fix is coordinated control. The units in a contained zone share the demand signal, commonly the rack inlet temperature or the hot-aisle return temperature for the zone, and they modulate together to hold it, splitting the load instead of competing for it. Many in-row product lines run a group or team control mode built for exactly this, where the units talk to each other or to a common controller and stage their fans and valves as one. Confirm the group control is configured and working, because a row of units in standalone mode is the default that ships, and standalone is the mode that fights.

What each unit modulates is the fan speed and the water valve, or the compressor staging on a DX unit, against the zone temperature. The sensor location decides everything. A sensor in a dead spot reads a temperature the racks never see, so the units chase a phantom. Verify the control sensors are where the design put them and reading the air the racks actually draw, then watch the units hold the zone as the load steps, not just at one quiet setpoint. The building management system ties the zone signals together and is where the coordination and the alarms live, so the BMS integration is a commissioning item in its own right.

Will an in-row unit make condensate, and how is it managed?

An in-row unit makes condensate whenever its coil runs below the dew point of the air passing through it, and the standard answer in a modern hall is to design so it mostly does not. Running the coil above the room dew point keeps the cooling sensible, meaning the unit cools the air without wringing water out of it, which is the preferred mode because the air in a data center carries the heat as temperature, not as moisture. A sensible-only design needs no drain at the unit, which is one less water source sitting next to the cabinets.

Holding the coil above dew point is why warm-water and warm chilled-water designs pair so well with in-row cooling and with the wide ASHRAE humidity band. The supply water can run warmer than a comfort-cooling chiller would make, the coil stays above the dew point, no condensation forms, and the economizer carries more hours of the year. The dew point, not the relative humidity, is the number that governs condensation, the same point the cooling pillar makes about the room as a whole.

Some units and some conditions will still condense, so the condensate management has to be real and tested where it exists. That means a drain pan that actually drains, a condensate pump that lifts the water to a drain where gravity will not, and an overflow or water sensor that alarms and, where the design calls for it, shuts the unit down before the pan overflows onto the floor of the row. Test the pump and the overflow by filling the pan, not by reading the schematic. A condensate pump that was never run is a flood waiting for the first humid day the economizer hands off to mechanical cooling.

The commissioning sequence for in-row cooling

In-row commissioning runs the same level structure as the rest of the mechanical plant, building from the installed component up to the row under load, and nothing functional gets tested until the static checks are signed. Component verification comes first: the units are the ones specified, set in the right positions in the row, piped or charged correctly, powered, and with the controls points checked. On a chilled water unit, the piping is hydrostatically tested and flushed before water touches a coil. On a DX unit, the refrigerant circuit is proven and charged to the manufacturer's figure.

Then the leak check, because water in the row is the failure that hurts most. With the loop pressurized and proven, you wet the leak sensors and confirm the alarm reaches the BMS and drives whatever response the design specifies. After that comes airflow and capacity verification, the test-and-balance work that proves each unit moves its design air and water and the zone delivers the cold air where the racks draw it. This is performed to the AABC or NEBB procedures, the same as the rest of the cooling balance, and it is where the gap between the design drawings and the as-built row gets found.

Functional performance testing then proves the units do what their sequence of operations says: the group control coordinates the row, the fans and valves modulate to hold the zone temperature, the units stage as load changes, and the condensate and leak responses fire on cue. The keystone is the integrated systems test under load, where the row is loaded with load banks or real load and put through the failure scenarios. The redundancy and failover test lives here, and so does the proof that the cooling rides through a power event, which on a real facility runs alongside the electrical integrated test because the cooling cannot survive a power blip the power side has not proven.

Verifying capacity, delta-T, and turndown under load

You cannot accept an in-row row on a cool, empty afternoon, because cooling only shows its true behavior when there is heat to remove. The load is supplied by load banks or by real IT load, placed in the racks to mimic the design heat distribution, and the units are brought to their design supply temperature and flow with the containment closed. A row that holds beautifully at ten percent load and falls apart at full load passed the wrong test, and full load is the only one the owner cares about.

The numbers you take are the capacity each unit actually delivers and the delta-T across the coil. A healthy in-row unit drawing genuinely hot return air shows a strong, stable delta-T, which is the signal that the air picked up a full load of heat before it reached the coil. A weak delta-T at the unit means the return came back cooler than design, which on an in-row layout almost always points back at the containment: the supply is short-circuiting around the racks and diluting the return before it reaches the unit. The delta-T at the unit is the tell that ties the cooling result to the air management, the same way the cooling pillar treats low delta-T as the signature of mixing.

Hold the row at design load long enough to see it stabilize, and map the rack inlet temperatures top to bottom across the zone while it runs. The unit can report a perfect supply temperature and still leave the top of a rack hot if the containment leaks or the air is short. The acceptance is the inlet map inside the envelope at full load, with the units holding it as a coordinated group, not the discharge temperature on the unit's own display.

Then test the turndown, because the load is not constant and in-row cooling is good at following it only when the controls behave. As the IT load drops, the units should ramp their EC fans and throttle their valves down to match, spending less fan and pump energy while still holding the inlet. The risk at low load is the units chasing each other into instability, the same fighting problem the controls section names, made worse because there is little heat to damp the swings. Step the load down, hold it, step it back up, and confirm the units track smoothly and the group control keeps them sharing the load. Watch the coil too: a unit running a colder coil at part load to satisfy a control loop can dip below the dew point and start condensing where it never did at full load, so confirm the coil stays above the dew point across the whole load range, not just at the design point.

How do you test in-row cooling redundancy and failover?

You test in-row redundancy by pulling a unit under load and confirming the row stays inside the envelope while the neighbors pick up its share. Redundancy in the row uses the same N notation as the rest of the plant. N is exactly the cooling the row needs. N+1 adds one more unit than the load requires, so any single unit can fail or be pulled for service and the remaining units still hold the row. In many in-row designs every unit runs together at part load, including the redundant one, which lowers stress and makes the handoff smoother when one drops.

The test is blunt and it is the one that matters most. With the row at design load and the containment closed, you fail a unit, by command, by killing its power, or by the method the test plan specifies, and you watch the rack inlet temperatures across the zone. The surviving units should ramp up, take the dropped unit's load, and hold the inlets inside the envelope through the transition. You time how fast the inlets recover and confirm they never cross the limit. A row that alarms or rides above the envelope when one unit drops is not N+1 in practice, whatever the nameplate count says.

The failure that hides here is the unit too far from the heat. In-row redundancy assumes the surviving units can reach the racks the failed unit was serving, but if the dropped unit sat at one end of a long row, the units at the other end may not move air far enough to cover its zone before the local racks heat up. That is the in-row version of the perimeter air-path problem, and it is exactly why you test the failover at full load with the actual unit positions, not on a spreadsheet that assumes the cooling is evenly reachable. Pull the unit, watch the inlets, record the recovery. That record is the redundancy.

Integration with the chilled water plant and the economizer

A chilled water in-row unit is only the last coil on a loop that runs back to a central plant, and the row cannot be accepted in isolation from what feeds it. The plant makes the chilled water, pumps move it through the loop to the in-row coils, and the warmed water returns to be cooled again, the same architecture the cooling pillar walks through for CRAH units. The in-row unit's design flow and temperature are a demand on that plant, and the plant has to deliver them at the same time to every row at full load, which is a different test from proving one row in isolation.

The economizer is where the energy savings live, and warm-water in-row designs are built to lean on it. When the outside air is cool enough, a waterside economizer makes chilled water with the cooling tower alone and lets the compressors idle. Because in-row units can run a warmer supply and still hold a dense rack, the design opens more hours of the year to free cooling, and every hour the economizer carries the load is an hour the chiller sits off. The changeover setpoints, the temperature at which the plant moves between mechanical and free cooling, are commissioning items, because they are where the operating efficiency is won or lost.

What this means for the in-row scope is that the row's functional test and the plant's functional test connect at the loop. A row that holds its inlets when the plant is making cold water on a calm afternoon still has to hold them when the economizer hands off, when a chiller stages, and when the loop temperature moves. The integrated test is what proves the row rides through those plant transitions, which is why the in-row units, the plant, and on a real facility the electrical side all get tested together rather than each signed off alone.

When does in-row run out and liquid take over?

In-row cooling runs out when a single rack pulls more heat than any volume of air can carry through it, which on the AI and GPU floor is arriving fast. Air-cooled in-row holds dense racks well into the tens of kilowatts when it is paired with tight containment, but the racks behind current AI training and inference hardware now pull 40, 80, and past 100 kW, and at those densities there is not enough air you can physically push through one cabinet to remove the heat no matter how cold or how close the unit. That is the ceiling, and it is a physics ceiling, not a product one.

In-row is the top of the air ladder, and the rung above it is liquid. Rear-door heat exchangers extend an air-cooled hall by pulling the heat out of the exhaust at the cabinet, and direct-to-chip liquid takes the high-density heat straight off the silicon while air still handles the rest of the rack. The liquid cooling guide covers the loop, the coolant, the flush, and the leak strategy that liquid demands. The point for the in-row commissioning agent is that most halls are now hybrid: dense liquid-cooled racks alongside conventional racks that in-row air still serves, often in the same room.

So in-row does not disappear when liquid arrives. It becomes the air half of a hybrid floor, cooling the memory, the drives, the power supplies, and the non-liquid racks that liquid never touches, while the cold plates handle the chips. A cold plate that runs perfectly does nothing for the components the air was supposed to cool, which means the in-row air path and the liquid loop get commissioned at the same racks. Knowing where the air job ends and the liquid job begins is what keeps a hybrid hall from a row that throttles because everyone assumed the liquid covered it.

Monitoring and the rack inlet temperature

Once the row is accepted, the number operations runs against is the rack inlet temperature, held inside the ASHRAE TC 9.9 recommended envelope, commonly given as 18 to 27 C (64.4 to 80.6 F) dry-bulb, with the actual equipment class and the current edition controlling the real limit. The cooling pillar carries the envelope and the A1 through A4 allowable classes in full. For in-row the point is narrower: the units are controlling to a zone temperature, so the monitoring has to confirm the inlets the racks actually see, not just the air the units report leaving the coil.

Inlet sensing is the monitoring that earns its place. A temperature probe at the inlet of representative racks, top, middle, and bottom across the zone, tells you whether the in-row units are holding the rack or whether a leak or a short-feed is letting the top of a rack drift up while the unit display reads fine. The vertical gradient is the same tell the containment guide names: clean at the floor and warm at the top means recirculation, not a cooling shortfall. Trend the inlets, not the room.

The unit's own telemetry rounds it out: supply and return temperature, fan speed, valve position, water or refrigerant condition, filter differential pressure, and every alarm it can raise. All of it has to reach the BMS and read correctly, because an alarm that lives only on the unit's local screen is an alarm nobody sees at three in the morning. The monitoring is what turns the commissioning baseline into something operations can trend against, so a warm rack next year can be checked against how the row was accepted instead of guessed at.

The upkeep after handoff

An in-row unit sitting in the row, breathing the room's air all day, comes with maintenance the owner takes on at turnover, and naming it in the commissioning deliverable is what keeps the row from degrading quietly. The filters are first. An in-row unit pulls a lot of air across a filter, and a loaded filter raises the pressure drop, cuts the airflow, and starves the row the same as a short feed would. The filter-change interval and the filter differential-pressure alarm setpoint belong in the turnover package, not in operations' guesswork.

The coil is next. A fouled coil loses capacity gradually, and on a chilled water unit the loss hides behind a valve that just opens further to compensate until it runs out of room. A baseline of the coil's clean performance at commissioning is what lets operations catch fouling before the row runs short. The condensate path, where one exists, needs the pan kept clear and the pump and overflow exercised, because a condensate pump that sits idle for a year is the one that fails the first time it is asked to lift water.

The fans round it out. EC fans are durable but not eternal, and a fan that drifts down in output drops the unit's contribution to the zone, which the surviving units cover until they cannot. The maintenance plan, the filter interval, the coil baseline, the condensate exercise, and the fan check, is part of what commissioning hands over. A row turned over without it is a row that will degrade until something throttles and nobody knows why.

Field example: a row that lost a unit and stayed up, after the fix

A colo row of eight chilled water in-row units, sold as N+1 for a row of dense cabinets, threw high-inlet alarms the first time a unit was pulled for a filter change, even though the design said any one unit could drop without consequence. The reflex was to call the units undersized. The numbers said the cooling was there and the air management was not.

With all eight running, the row held fine. Pull the unit at the far end and the three racks nearest it climbed past the envelope within minutes, while the units at the other end of the row barely changed their output. The cause was two ordinary things. The units were running in standalone mode, each on its own sensor, so the far units never saw the demand from the racks that lost their local cooling, and the containment had a gap at the end of the row that let the now-uncooled racks pull exhaust the moment their unit stopped. The redundant capacity existed; it could not reach the racks that needed it.

Setting the units to group control on the shared zone signal and sealing the end-of-row gap changed the result. Pull the same unit again and the remaining units ramped together, took its share, and held the inlets inside the envelope through the transition, with the worst rack peaking a couple of degrees and recovering. No cooling was added. The row had been N+1 on the nameplate and standalone in the controller, which is not the same thing as redundant.

What to document

An in-row row that was proven but never documented hands operations a zone with no baseline, and the first warm rack becomes a guessing game. The record is what tells the next engineer whether a reading is a new problem or how the row has always run, and what the row was accepted at. Capture it per zone and per unit, not as a single room number, because in-row cooling is a per-zone job and the average hides the row that fails.

Record the unit identity and the cooling medium, the design and as-balanced air and water flow, the supply and return temperatures and the delta-T per unit, the rack inlet map top to bottom across the zone, the redundancy scheme and the failover test result, the leak-detection test, the condensate and dew point check, the control mode and the coordinated-control verification, and the load the row was tested at. The failover result and the inlet map carry the most weight later, because they are the two questions you cannot re-answer once the hall is in production.

Common mistakes

• Running in-row units with no containment, so they draw mixed air, the return sags, and the density never shows up.
• Leaving the units in standalone control, where each chases its own sensor and the row hunts and fights itself.
• Sizing the in-row capacity on the hall average instead of the load in each zone, starving the dense row.
• Accepting the row at low load and never testing it at the load it will actually carry.
• Calling a row N+1 on the nameplate without pulling a unit under load to prove the neighbors hold the inlets.
• Ignoring condensate and dew point, then flooding a cabinet the first time a coil dips below the dew point.
• Skipping the chilled water hydrostatic test and flush, then finding the leak next to powered cabinets.
• Installing leak detection but never wetting a sensor to prove the alarm and the shutdown actually fire.
• Judging the row by the unit's discharge display instead of mapping the rack inlets top to bottom.
• Placing the redundant unit where the survivors cannot reach its racks before they heat up.

Field checklist

Standards and references

The thermal target the in-row row is held to comes from ASHRAE Technical Committee 9.9 and its Thermal Guidelines for Data Processing Environments, which set the recommended and allowable server-inlet envelopes and the A1 through A4 equipment classes. For chilled water in-row units, the warm-water and free-cooling logic ties to the same guidance, and the cooling pillar carries the envelope and the architecture in full. The energy side is ASHRAE Standard 90.4 for data centers, which bounds the mechanical and electrical overhead by climate zone.

The in-row unit's capacity, the design flow, the refrigerant charge on a DX unit, and the coordinated-control behavior are governed first by the manufacturer's specification and the sequence of operations, which override any rule of thumb where they are stricter. The density crossover between perimeter and in-row, and the figures for fan-energy savings, are design and vendor numbers, not code lines, so treat them as a band and confirm them against the actual product and project. The chilled water piping test falls under the project specification and the applicable piping code; the test and balance is performed to the AABC or NEBB procedures.

The commissioning process framework draws on the ASHRAE commissioning guidance, commonly Guideline 0, alongside the Building Commissioning Association best practices. Where a facility is chasing a tier, the Uptime Institute Tier standards drive the witnessed redundancy and integrated-test demonstrations, and TIA-942 is the broader telecommunications infrastructure standard for data centers, including environmental and redundancy provisions. Edition numbers and the specific values move between cycles, so confirm the current edition and the actual figures against the published documents and the equipment manufacturer before citing them on a submittal.

Units, terms, and acronyms

In-row cooling carries vocabulary from HVAC, from the IT side, and from commissioning, and the same idea reads differently across a unit submittal, a TAB report, and a Tier document. The terms below travel across the whole in-row scope.

In-row / close-coupled

A cooling unit set in the row between cabinets, drawing hot-aisle return and supplying the cold aisle over a short air path

CRAH / CRAC

Computer room air handler (chilled water coil) or air conditioner (DX with a compressor); in-row units come in both forms

Delta-T

The temperature rise across the unit coil; a strong, stable delta-T means the unit is drawing genuinely hot return air

CFM / kW

Airflow per heat load; roughly 160 CFM per kW at a 20 F rise sets how much air a row needs

EC fan

Electronically commutated fan that modulates speed smoothly, the part that lets in-row turn down and save fan energy

Group / team control

The coordinated mode where in-row units in a zone share a demand signal so they split load instead of fighting

Dew point

The temperature at which air condenses; holding the coil above it keeps cooling sensible and avoids condensate

N+1

One more cooling unit than the row needs, so any single unit can drop and the survivors still hold the inlets

Rack inlet temperature

The air temperature at the server intake, the number judged against the ASHRAE TC 9.9 envelope, not the room average

Sensible cooling

Cooling that lowers air temperature without removing moisture, the mode an in-row unit runs when the coil stays above dew point

Related tools

Calculators and readiness checks for this work

Compare your options

FAQ

People also ask

• What is bypass airflow in a data center?
• What is direct-to-chip liquid cooling?
• What is a CDU in a data center?
• Why do data centers need liquid-cooling leak detection?
• What is the difference between hot-aisle and cold-aisle containment?