CRAC vs CRAH: which failure buys you time, and which one takes the whole hall
Every comparison of CRAC and CRAH units ends with one sentence on redundancy: CRACs are independent, CRAHs depend on the plant. Taken one failure at a time, the independence belongs to the heat-rejection topology, not the unit, and a glycol CRAC shares a dry cooler exactly as a CRAH shares a chiller.
By Jeel Patel, Founder at HVAC Software
At 14:20 on a July afternoon CRAC-04 trips on high head pressure, and eleven minutes later CRAC-02 does the same. Both units are air-cooled, both condensers sit on the same roof, and the hall's N+1 was written on the assumption that CRAC failures arrive one at a time. In the chilled-water hall next door nothing has tripped, and this scene is a composite.
A CRAC is a computer room air conditioner with its own compressor and refrigerant coil, and a CRAH is a computer room air handler with a fan and a chilled-water coil fed by a central chiller plant. Which failure buys you time depends less on which unit you own than on what the unit shares. A CRAC's own failures are bounded to one unit and covered by the spare, while a CRAH's shared failures, in the chiller, the pumps or the loop, take every unit on the loop at once but leave minutes of cold water in the pipes. A glycol or water-cooled CRAC shares a dry cooler or tower exactly as a CRAH shares a chiller.
- Independence is a property of the heat-rejection path, not the unit type. An air-cooled CRAC is one unit per condenser. A glycol CRAC shares a dry cooler and pump package with several others, and a Dual-Cool CRAC borrows the chilled-water plant.
- A CRAH's own failures are few: the fan, the control valve and a leak, each bounded to one unit and each with the OEM's own protective stop. Its shared failures are the plant's, and those consume every unit's cover at once.
- Time after a shared failure is set by what is cold near the coil: pipe water for a CRAH, nothing for a DX unit until its compressor restarts, which Schneider puts at several minutes. The window is ASHRAE's 5 °C in any 15 minutes.
- The one repair a CRAC has and a CRAH never has, opening the refrigerant circuit, is gated on an EPA Section 608 card, and from 1 January 2027 a new CRAC holds a mildly flammable A2L refrigerant in the hall.
What is the difference between a CRAC and a CRAH, in one table?
Schneider's taxonomy counts 13 fundamental heat removal methods between the hall and the outdoors, and the two units on this page sit at opposite ends of it. In a CRAC the refrigeration cycle is in the room, split between the unit and its condenser or contained entirely inside a glycol-cooled cabinet. In a CRAH the refrigeration cycle has been moved out of the room into a chiller that produces water at about 8 to 15 °C, and the unit in the hall is a fan, a coil and a valve.
| CRAC | CRAH | What it shares with the units beside it | |
|---|---|---|---|
| What cools the air | A refrigerant coil fed by the unit's own compressor | A chilled-water coil fed by the plant | CRAC: nothing, on the air-cooled kind. CRAH: the water |
| Where the refrigeration cycle sits | In the unit, or half in the unit and half in the outdoor condenser | In the chiller plant, outside the hall | CRAH: the whole cycle is shared by every unit on the loop |
| How the heat leaves | One air-cooled condenser per unit, or a dry cooler and pump package serving several, or a cooling tower | The chilled-water loop to the chiller, then the chiller's own condenser water or air | CRAC: the dry cooler, its pump and the tower are shared. CRAH: everything upstream of the valve |
| Moving parts in the hall | Compressor, fan, expansion valve, condensate pump, humidifier | Fan, control valve and actuator, condensate pump, humidifier | Neither shares these. They are the bounded failures |
| Liquids in the hall | Refrigerant, 12 to 23 lb per unit on one Liebert family, plus condensate | Chilled water at plant pressure, plus condensate | A CRAH leak is water; a CRAC leak is a regulated gas |
| Where each is usual | Rooms of 7 to 200 kW, in Schneider's bands | Halls of 200 kW and up | The shared plant is what makes the CRAH cheaper per kilowatt at scale, and what makes its failures travel |
- DX
- Direct expansion: refrigerant evaporates in the coil the air passes over. Schneider notes any system with refrigerant and an evaporator coil is DX, not only the air-cooled kind.
- Air-cooled CRAC
- The unit indoors and a condenser outdoors, joined by field-installed refrigerant lines that cannot run far. One condenser serves one unit.
- Glycol-cooled CRAC
- The whole refrigeration cycle sealed inside the cabinet, rejecting heat to a glycol loop, a pump package and an outdoor dry cooler that can serve several units.
- Water-cooled CRAC
- As glycol-cooled, but rejecting heat to condenser water piped from a cooling tower.
- Dual-Cool and GLYCOOL
- A CRAC with a second coil: chilled water from the plant when it is available, or cold glycol when the outdoor temperature allows, reducing or eliminating compressor operation.
- Head pressure
- The refrigerant pressure on the compressor's discharge side. It rises with outdoor temperature and falls with condenser airflow, and the unit's controls and safeties act on it.
- Two-circuit unit
- A CRAC with two independent refrigeration circuits, each with its own compressor, low-pressure cut-out and heat-rejection interlock, so that one circuit tripping leaves the other running.
- Control valve
- The 2-way or 3-way valve that meters chilled water through a CRAH coil to match the load. The one part of a CRAH that decides how much of the plant's cold it takes.
The size bands are Schneider's and the reasons behind them are the ones every page repeats. Chilled-water CRAH units cost less, contain fewer parts, and have greater heat removal capacity than CRAC units with the same footprint, and chilled-water piping can run the length of a building from one plant. The same paper lists the one disadvantage the rest of this page turns on: a chilled-water system introduces an additional source of liquid into the IT environment.
Why does the heat-rejection topology decide how far a failure spreads?
Because the unit is only the last metre of the path the heat takes, and a failure anywhere on that path reaches every unit that shares it. For an air-cooled CRAC the path is private: Schneider states that multiple computer room air conditioners cannot be attached to a single air-cooled condenser, so a condenser fan that stops takes exactly one unit with it. That is the whole basis of the sentence every comparison page carries, and it is true of that one topology.
It is not true of the others. The same paper says a glycol loop can service several CRAC units from one dry cooler and pump package, which means a pump that stops or a dry cooler that loses its fans takes every CRAC on that loop, and a water-cooled unit is piped to a cooling tower it shares in the same way. Vertiv's Liebert DS manual describes the Dual-Cool option as a compressorised unit that adds a second cooling coil that is connected to a source of chilled water, which gives the unit a second source and the plant's shared failures at the same time.
CRACs are independent and CRAHs depend on the plant is a statement about one condenser per unit. Change the heat rejection and the CRAC inherits the CRAH's failure pattern without anyone noticing.
The consequence for redundancy is arithmetic rather than opinion. A bounded failure consumes one unit's cover, so N+1 at the unit level survives it. A shared failure consumes every unit's cover on that source at once, so the spare that matters is the spare on the source, which is the distinction the N+1 versus 2N guide draws for the plant and this page draws for the unit.
Which CRAC failures stay inside one unit, and which do not?
The table takes the failures a DX unit has of its own, with what one OEM's manual says the unit does when each occurs. The Liebert DS family runs from 28 to 105 kW on two refrigeration circuits, and the figures are that family's figures, presented as such. The spread column is the one that decides whether the spare covers you.
| Failure | What the unit does, per the Liebert DS manual | Spread | Who repairs it |
|---|---|---|---|
| High head pressure trip | The condenser fan is speed-controlled on head pressure, starting to rotate at 190 psig and reaching full speed at 250 psig, with a discharge relief valve rated at 480 psig maximum. A trip on one circuit leaves the other circuit running on a two-circuit unit | Bounded to one unit. The cause may not be: on a hot afternoon every air-cooled condenser on the roof sees the same ambient and the same head-pressure rise, which is an inference from the control physics, not a figure from the manual | A qualified technician to find the cause and reset. An EPA 608 card only if the circuit is opened |
| Low suction pressure, low charge | A minimum of 20 psig must be established and maintained for the compressor to operate, and each circuit's low-pressure cut-out is recorded at commissioning | Bounded to one unit. A charge loss on an HFC unit of 15 lb or more also starts the leak-repair clock the PM checklist explains | EPA 608 Type II to open the circuit and recharge |
| Compressor burnout | If a burnout has occurred, a full system clean-out is required, or compressor and system problems will continue | Bounded to one unit, and long. The unit is out for a replacement and a clean-out, not a reset | EPA 608 Type II, and the OEM's replacement procedure |
| Condenser fan or coil failure, air-cooled | The unit's head pressure climbs with no airflow across its condenser and the circuit trips as above | Bounded to one unit, because one condenser serves one unit | Mechanical and electrical; no card unless the circuit is opened |
| Dry cooler, glycol pump or tower loss | The glycol or water carrying the heat stops moving and the units on that loop trip on head pressure together | Shared by every CRAC on the loop. This is the CRAH failure pattern inside the CRAC family | Mechanical; the pump package or the tower, not the unit |
| Fan motor fault | Vertiv's EC fan protection stops the motor electronically on overtemperature, rotor fault, locked rotor or phase failure, with no automatic restart | Bounded to one unit, and hard: the unit moves no air until a person cycles it | A properly trained and qualified technician, in the manual's words |
| Clogged filter | A differential switch raises the Change Filter alarm while the unit keeps running on reduced airflow | Bounded to one unit, and slow | Anyone on the PM list |
| Utility power loss | Every unit on the feed stops together and restarts on generator | Shared by the hall. The failure-time ranking has the restart chain | Nobody; the restart is the plant's |
Two rows deserve their sources spelled out. The 20 psig floor, the 190 and 250 psig fan-control points and the 480 psig relief rating are all in the DS manual's charging and head-pressure control procedures, which also describe fan-speed control that keeps the unit operating to 0 °F and a flooded-condenser system for operation to −30 °F. The manual is written for the cold end of the ambient range, and the hot end, where every condenser on a roof climbs together, is the one this page flags.
Which CRAH failures stay inside one unit, and which take the hall?
A CRAH has fewer failures of its own because it has fewer parts, and the parts it has come with protective behaviour written into the OEM guide. The table uses Vertiv's Liebert CW installer guide for the 38 to 181 kW range and its PEX+ chilled-water manual for the valve. Read the spread column against the one above.
| Failure | What the unit does, per the Liebert CW and PEX+ guides | Spread | Who repairs it |
|---|---|---|---|
| Chilled-water flow stops at the plant | The coil goes on cooling with the water already in it, then warms. A flow switch on the unit activates the warning system | Shared by every CRAH on the loop, at once. The cold in the pipes is the only buffer | Mechanical, at the pump or the loop; nothing on the unit is broken |
| Chilled-water supply warms, chiller fault | The valve opens further to hold setpoint until it is fully open, then the unit's capacity falls with the water temperature | Shared by every coil on the loop. Every unit loses capacity together, without a single unit alarm at first | The chiller technician, who at a low-pressure chiller holds a 608 Type III card |
| Control valve or actuator fails | The 2-way or 3-way high-pressure valve stops metering. The PEX+ manual describes the valve as a body, connector and actuator that opens further as the load rises | Bounded to one unit. The coil is still cold; it is either unmetered or starved | Controls or mechanical; the actuator is a replaceable part |
| Fan motor fault | The EC fan's monitoring stops the motor electronically on overtemperature, rotor position, locked rotor, undervoltage or phase failure, and the guide states there is no automatic restart and power must be off for at least 20 seconds | Bounded to one unit, and hard until a person cycles it | Only a properly trained and qualified technician should open the motor |
| Water leak inside the unit | The guide requires monitored leak detection that closes spring-return shutoff valves on the supply and return immediately on a leak signal | Bounded to one unit by the unit's own protection, which removes it from service | Mechanical; the leak, then the valves |
| Strainer blocked, flow switch trips | Flow through the coil falls and the switch raises the warning | Bounded to one unit | Mechanical, on the unit's own piping |
| Clogged filter | The Clogged Filter or Change Filter alarm, unit still running | Bounded to one unit, and slow | Anyone on the PM list |
| Utility power loss | Every fan and every pump on the feed stops together | Shared by the hall, with the restart chain and stored-water figures in the ranking guide | Nobody; the restart is the plant's |
The leak row is worth reading twice, because it is where the two units differ most. Vertiv's guide tells the installer to fit shutoff valves in the supply and return water lines that automatically close if water is detected by the leak detection system, spring-return and rated above supply pressure. A CRAH's worst private failure therefore ends with the unit isolating itself, and a CRAC's worst private failure, a burnout, ends with a clean-out.
The fan row is the same on both units, because both families use the same EC fan protection. The guide's wording is that with any of these failures the motor stops electronically and there is no automatic restart, so a fan fault at 03:14 is a unit that stays off until someone is standing at it, whatever it is cooled by.
How does one failure travel from the source to the rack inlet?
Every failure in the two tables sits somewhere on one path, and the path is the same for both units. Above the line is what the unit shares with its neighbours, below it is what the unit owns, and the unit type moves components across the line without changing the line. The diagram draws it once for both.
Two things follow from the picture. The first is that a CRAH's simplicity is real but local: everything that can fail below the line on a CRAH is a fan, a valve or a leak, each with a protective stop, and everything else that can fail has been moved up the path where it is shared. The second is that the same move is available to a CRAC, and a site that piped its CRACs to one dry cooler to save roof space made it without changing the word on the drawing.
How long does each failure leave you inside the ASHRAE window?
The window is the same for every row. ASHRAE's 2021 thermal guidelines allow ITE inlet temperature to change by no more than 5 °C in any 15-minute period and 20 °C in an hour, measured as a change within the period rather than a rate, and the recommended envelope is 18 to 27 °C. What differs by row is what is still cold near the coil when the failure lands.
| Failure class | What carries the load | Time, as the source states it | Source |
|---|---|---|---|
| Bounded failure on any unit, spare running | The N+1 unit, immediately, if it is actually spare and not already carrying a unit on a permit | The window is the spare's, and the redundancy guide shows how often it is not spare | Definition of N+1 |
| Bounded failure, spare in standby | The standby unit once its controller starts it | No figure is quoted here; it is the controller's start sequence and the site's own test record | Not sourced; site-specific |
| CRAH loop loses its chiller, pumps still running | The water already in the pipes, then nothing until the chiller is back | Chiller restart typically 10 to 15 minutes on a standard machine, 4 to 5 on a quick-start design | Schneider WP 179 |
| CRAH with stored chilled water, fans and pumps on UPS | The storage tank through the restart | Chilled-water storage, if available, can be utilised immediately by CRAHs with only minimal backup power | Schneider WP 179 |
| Glycol or water-cooled CRAC loses power or its loop | Nothing until the unit's compressor restarts, because the fluid is useless to a DX coil until then | The restart may take several minutes, and stored fluid cannot be used until the CRAC units restart | Schneider WP 179 |
| Air-cooled CRAC loses power | Nothing until the unit and its condenser both have power and the compressor restarts | The least emergency-performance potential, because the entire system including the outdoor heat rejection must be on backup power | Schneider WP 179 |
| Any unit, fan fault | The neighbours, through the room's own mixing | Until a person cycles the power, with a 20-second minimum off time | Vertiv Liebert CW guide |
The DX rows come from Schneider's outage paper, which states that chilled-water systems can offer advantages over fluid and air-cooled DX systems when facing a loss of primary power for two reasons: a CRAH has no compressor to power, and any stored chilled water can be used at once. The same page ranks air-cooled DX last on emergency performance, and it adds that putting glycol pumps on UPS gains nothing for a DX unit without a free-cooling coil, because the fluid cannot be used until the compressor is back.
The bounded rows have no physics in them, and that is the point. A CRAC losing a compressor at 14:20 costs the hall nothing if CRAC-06 is spare and running, and costs it the whole window if CRAC-06 is the unit the permit isolated at 08:30. The time a bounded failure leaves you is a fact about the spare, and the ranking guide's method for finding which failure ends your hall first is the way to put a number on it.
Who is allowed to repair each failure?
The credential follows the failure, and only one credential separates the two units. Federal rule 40 CFR 82.161 requires that anyone who could reasonably be expected to violate the integrity of the refrigerant circuit during maintenance, service, repair or disposal must pass a certification exam, with Type II for medium-, high- and very high-pressure appliances, which is where a CRAC's circuit sits. A CRAH has no refrigerant in the hall, so the card never applies to it, and it applies at the chiller instead.
| Repair | CRAC | CRAH | Rule or source |
|---|---|---|---|
| Open the refrigerant circuit: recharge, leak repair, compressor replacement | EPA Section 608 Type II or Universal | Not applicable in the hall; Type III at a low-pressure chiller | 40 CFR 82.161(a)(1)(ii) and (iii) |
| Fan motor fault | A properly trained and qualified technician | The same, in the same words | Vertiv EC fan section |
| Chilled-water valve, actuator, strainer or leak | Dual-Cool units only | Mechanical or controls; no refrigerant card | Vertiv CW and PEX+ guides |
| Work inside the live panel | OSHA's qualified person | The same | 29 CFR 1910.399, per the qualification guide |
| Isolate the unit for any of the above | A lockout-tagout authorised employee | The same | 29 CFR 1910.147, per the same guide |
| Buy a replacement unit after 1 January 2027 | A refrigerant at or under 700 GWP, which rules out R-410A at 2,088 and leaves A2L options such as R-454B at 465 and R-32 at 675 | No refrigerant decision in the hall | EPA technology-transitions sector table and GWP table |
The last row is the one that is about to change the comparison. EPA's technology-transitions rule restricts data centers, computer room air conditioning, and information technology equipment cooling to refrigerants of 700 GWP or less for manufacture, import and installation from 1 January 2027, and the agency's GWP reference table puts R-410A at 2,088, R-454B at 465 and R-32 at 675. EPA's SNAP listing classes both R-454B and R-32 as A2L, the mildly flammable class under ASHRAE 34, so a CRAC bought from 2027 puts a flammable refrigerant in an occupied data hall, which is a design and a qualification question a chilled-water hall never has to ask.
What do the BMS, DCIM and CMMS already hold about the unit?
Each system holds one input to the spread question and none holds the answer. The unit controller knows the unit tripped, the BMS knows which points moved, the DCIM knows which racks the unit serves, and the CMMS knows which technician holds which card. What none of them holds is the line in the diagram: which source this unit shares, and therefore how many units the alarm is really about.
| System | What it holds about the failure | What it does not hold |
|---|---|---|
| Unit controller | The alarm itself, the head pressure, the fan state, the valve position, and a common-alarm contact closed on any alarm, which on the Liebert DS is terminals 75 and 76 | Whether the unit next to it is about to raise the same alarm for the same cause |
| BMS or BAS | Supply and return air, chilled-water supply temperature, pump and chiller status, the setpoint | Which units share which dry cooler or loop. That is a drawing, not a point |
| DCIM | The racks each unit serves, the hall's density, the environmental history | The heat-rejection topology behind the unit |
| CMMS or EAM | The asset, its PM history, the OEM manual as an attachment, the technician's 608 type | The spread of the failure, so the work order for CRAC-04 does not know CRAC-02 is next |
| The mechanical drawing | The topology: which condenser, which loop, which tower | The live state. It was drawn once and the permit that isolated CRAC-06 is not on it |
CRAH-07 supply air above limit
- ✓Confirm redundancy state at panel
- ✓Isolate per LOTO — CHW-B branch
- Inspect valve actuator travel
- Verify supply air returns below 24 °C
- Restore N+1 and record final state
How to find out which of your units' failures are shared this week
The exercise is an inventory of sources, not a fault-finding procedure. For each hall unit, write down what is above the line and who else is on it, and the two failure tables above fill themselves in for your plant. The order below produces one page per hall that the shift and the on-call engineer can both read.
- List every hall unit with its type as built, not as labelled: air-cooled DX, glycol-cooled DX, water-cooled DX, Dual-Cool, or chilled-water CRAH. The nameplate and the piping decide, not the schedule.
- For each unit, name its heat-rejection path component by component: the condenser, or the dry cooler and pump package, or the tower, or the chiller plant and loop. Mark each component as private to the unit or shared, and list the units that share it.
- Count the units behind every shared component. A dry cooler with four CRACs on it is a four-unit failure, and a chiller plant with twelve CRAHs on it is a twelve-unit failure, whatever the unit-level N+1 says.
- Check where the spare actually is. If N+1 was counted in units, confirm there is also a spare on every shared source, and record which permits consume which spare, per the redundancy guide's live-state ledger.
- Copy the OEM's protective behaviours into the asset record: the fan protection that does not restart, the leak detection that closes the shutoff valves, the pressure cut-outs per circuit, and the burnout clean-out requirement. These are the failure rows for that model.
- For air-cooled units, write the condenser's rated ambient beside the site's design-day temperature, and treat a hot afternoon as a shared cause for units that share a roof, even though each trip is bounded.
- Put the credential beside each failure row: 608 Type II for anything that opens a CRAC circuit, Type III at the chiller, qualified person for the panel, and the LOTO procedure for the isolation, with the names on shift who hold each.
- For every shared row, write the window from what is cold near the coil, using the ranking guide's method, and for every bounded row write the spare's name and whether it is running or in standby.
The output is the missing column of the mechanical drawing: the spread of each failure, in units, with the spare and the responder beside it. It is also the answer to the question the composite scene opened with, because CRAC-02 following CRAC-04 was not two bounded failures but one shared cause, and a hall that had written the roof down as shared would have moved load before the second trip.
What this cannot do
A comparison assembled from two OEM families' manuals, two Schneider white papers, the ASHRAE reference card and three EPA pages is not a failure analysis of your plant. The Liebert DS figures are one product family's figures on R-407C, the CW and PEX+ behaviour is Vertiv's, and the restart times are Schneider's industry-typical values in a modelled hall. The AHRI rating standard for these units could not be opened for this page and is not cited. Every figure carries its source on the sentence that makes it.
Nothing here predicts a trip or diagnoses one. A decision layer that holds the spread can say that CRAC-04's high head pressure is on a roof shared with three other condensers at 36 °C ambient, that CRAC-06 is on a permit, and that the repair needs a Type II card the technician on shift does not hold. It cannot say the compressor will fail, and it does not reset the fan, open a valve, start a compressor or command the dry cooler. A certified technician opens the circuit, a qualified person opens the motor, and a named person decides which unit sheds load.
Which guides sit beside this one?
- Cooling systems ranked by failure time: the power-loss restart chain and the stored-water figures this page refers to rather than repeats.
- Cooling redundancy, N+1 versus 2N: the arithmetic of the spare a bounded failure consumes, and how a shared failure consumes them all.
- CRAC and CRAH preventive maintenance checklist: the leak-repair rule a low-charge row starts, and the permit that takes a unit out of rotation.
- Critical environment technician qualification: the credential ledger behind the who-repairs-it table.
- MTBF on cooling equipment: the one published field failure rate for a CRAH, and why none exists for a DX unit in the same set.
Answered
Is a CRAH more reliable than a CRAC?
In the unit, usually, and in the hall, only if the plant is. A CRAH has fewer parts, and the one published field set, Cheung and Wang's survey used in the [MTBF guide](/data-center/mtbf-on-cooling-equipment-what-it-predicts/), gives a chilled-water hall unit an implied MTBF near twelve years with no DX unit in the same table. Every failure the CRAH does not have has moved to the chiller, the pumps and the loop, where one failure takes every unit at once.
What happens to CRAH units when the chiller fails?
Every CRAH on that loop loses its cold source at the same moment, and the water already in the pipes is the only buffer. Schneider's outage paper puts a standard chiller restart at 10 to 15 minutes and a quick-start design at 4 to 5, and states that stored chilled water can be used immediately by CRAHs with minimal backup power. Without storage, the coils warm as the pipe water does, all together.
What happens when a CRAC compressor fails?
That one unit stops cooling, and on a two-circuit unit one circuit can keep running. The Liebert DS manual requires a full system clean-out after a burnout, so the repair is a replacement rather than a reset, and opening the circuit needs an EPA 608 Type II technician. The spare unit carries the load in the meantime, if the spare is running and not already on a permit.
Are CRAC units really independent of each other?
Only the air-cooled kind, and only in the unit. Schneider states that multiple computer room air conditioners cannot share one air-cooled condenser, so that topology is one unit per condenser. Glycol-cooled CRACs share a dry cooler and pump package, water-cooled CRACs share a tower, and a Dual-Cool unit borrows the chilled-water plant. Every unit on a shared component fails with it, and a hot roof is a shared cause even for air-cooled units.
Can a CRAC unit use chilled water as a backup?
Yes, on units built for it. Vertiv's Dual-Cool option adds a second coil connected to a source of chilled water, used when it is available to reduce compressor operation, and the GLYCOOL variant adds a coil fed by cold glycol when the outdoor temperature allows. Schneider calls the same idea a multi-cool option. The second coil is a second source, and it also gives the unit the plant's shared failures.
Who can repair a CRAC versus a CRAH?
The refrigerant circuit is the difference. Under 40 CFR 82.161 anyone who could reasonably be expected to violate the integrity of a CRAC's circuit must hold an EPA Section 608 card, Type II for its pressure class. A CRAH holds no refrigerant in the hall, so its fan, valve, strainer and leak repairs need a trained and qualified technician, OSHA's qualified person for the panel and a LOTO authorised employee for the isolation, but no 608 card.
Will new CRAC units use a flammable refrigerant?
From 1 January 2027, in the United States, they will use one at or under 700 GWP. EPA's technology-transitions table applies that limit to data centers, computer room air conditioning and IT equipment cooling for manufacture, import and installation, which rules out R-410A at 2,088. The compliant options EPA's GWP table lists, R-454B at 465 and R-32 at 675, are both classed A2L, mildly flammable, in EPA's SNAP listing.