Datacenter
Data center delta-T and return temperature management field guide
Read the air-side delta-T, find the bypass it exposes, and raise and match the rise across the IT and the cooling units to free the capacity already on the floor.
Direct answer
Delta-T in a data hall is the temperature rise of the air as it passes through the IT equipment, and that same rise should appear as return minus supply at the cooling unit. A low delta-T usually means cold supply is bypassing the servers, so units move huge airflow for little heat and waste capacity.
Key takeaways
- A low delta-T at the cooling units almost always means bypass air: cold supply short-circuits back to the return without passing through a server.
- Typical IT delta-T runs roughly 15 to 25 F (about 8 to 14 C), commonly designed around a 20 F rise for mixed rack-mount load.
- RTI equals unit delta-T divided by IT delta-T; 100 percent is the target, below 100 means bypass, above 100 means recirculation.
- Manage to the rack intake against the ASHRAE TC 9.9 envelope (near 18 to 27 C, about 64 to 80 F), not the room average.
- Fix airflow before raising supply temperature; bypass can reach 50 to 80 percent of supply on floors with unsealed openings.
Delta-T in the data hall, and why a low number is the warning
Delta-T in a data hall is the temperature rise of the air as it passes through the IT equipment. Cold air enters the server at the front, picks up the heat the chips throw off, and leaves the back warmer. That rise is the delta-T, and the same rise should show up at the cooling unit as the return temperature minus the supply temperature. When the air does its job, the two match the load.
The number that fools people is the low one. A low delta-T feels safe, because low usually means cool, and cool feels like margin. It is the opposite. A low delta-T at the cooling units means the return air is barely warmer than the supply, which means the cold air came back without ever climbing through a server. The unit is moving a wall of air and removing almost no heat.
Read the delta-T as a leak gauge, not a comfort reading. It is the single cheapest air-side number on the floor, it ties straight to capacity and energy, and it points at a specific failure when it drops. The rest of this guide is how to read it, what a low one is telling you, and how to raise and match it so the cooling you already own actually carries the load.
Why a low delta-T quietly wastes your cooling
A low delta-T costs money in two directions at once, and neither one shows up as an alarm. First, fan energy. A unit reading a low return is removing little heat per unit of air, so it has to move far more air to reject the load, and fan power climbs with the cube of airflow. You are paying to push a huge volume of air that does almost no work.
Second, capacity. A cooling unit is rated at a return temperature, and its real output rises as the return gets warmer. Starve it of return heat and the same unit delivers a fraction of its nameplate. The floor reads short on cooling while the capacity sits stranded inside units that cannot see enough heat to earn their rating.
This is the trap that drives the wrong purchase. The hall has a hot rack, the cooling looks maxed, so someone scopes another unit. The real problem was a delta-T of 5 F where the design wanted 20 F, and the cure was sealing air paths, not buying tonnage. The delta-T is measurable, it is on the floor today, and it tells you which problem you actually have before you spend.
What is delta-T in a data center?
There are two delta-Ts that matter on the air side, and a healthy floor keeps them close. The first is the rise across the IT: the inlet temperature at the front of the rack subtracted from the exhaust at the back. The second is the rise across the cooling coil: the return temperature into the unit minus the supply temperature off the coil. On a tight floor those two are nearly the same number, because the air that warmed up in the servers is the same air that comes back to the units.
When they diverge, the gap names the fault. If the unit delta-T is well below the IT delta-T, supply air is reaching the return without passing through a server, which is bypass. If the inlets read hotter than the supply you set, exhaust is leaking back to the front, which is recirculation. The two delta-Ts read against each other are a diagnosis, not just a pair of readings.
Both should track the load. Heat, airflow, and temperature rise are locked together: at sea level the rise in degrees F is roughly 3.1 times the watts divided by the CFM. Push the design airflow through the design load and you get the design rise. Get a rise far off that, with the load real and steady, and air is going somewhere it should not.
The Upsite framing of four delta-Ts, across the IT, across the unit, and the two negative ones that signal mixing, is a useful way to keep them straight. The CRAC and CRAH static pressure guide covers the unit-side setup that produces the rise; this guide is about reading the rise and managing to it.
The IT delta-T: the rise across the server
The IT delta-T is the rise the gear itself creates, and it is set by the equipment and its load, not by the room. A server pulls a fixed mass of air across its heat sinks and dumps it out the back warmer. The harder it works, the more heat per unit of air, and the larger the rise, up to the limit of its own fans.
The typical rise across IT equipment runs roughly 15 to 25 F, about 8 to 14 C, commonly designed around a 20 F rise for mixed rack-mount load, but treat that as a starting figure and confirm it against the actual gear. A dense compute rack under load can run a much larger rise, and the inlet-to-exhaust spread measured top of rack to bottom can read higher still because the top of the cabinet sees hotter air. The rack delta-T and the per-server delta-T are not the same number.
What you do not get to choose is the IT delta-T directly. It is what the load and the equipment produce. What you do control is whether the cooling side matches it. The whole air-side job is delivering the right volume of cold air to the inlet so the gear can make its design rise, then catching that warmed air and carrying it back to the units without diluting it.
The unit delta-T: return minus supply at the cooling unit
The unit delta-T is the return temperature into the CRAC or CRAH minus the supply temperature off its coil. It is the number on the unit's own controller, and it is the one operators watch, because it is easy to read and it reports on the whole air path feeding that unit. A unit holding a return near its design and a steady supply is seeing the heat it was sized to reject.
A low return temperature is a low unit delta-T, and it is the most common air-side complaint there is. The supply might be a perfect 65 F off the coil, but if the return comes back at 70 F, the unit has a 5 F delta-T and is doing a fraction of its work. The cooling capacity did not vanish. The heat never reached the coil to be removed.
Read the unit delta-T against the IT delta-T, never alone. A unit at 8 F return rise on a floor whose gear is making a 20 F rise is telling you that more than half the air it moves never touched a server. The CRAC and CRAH static pressure guide covers how the unit is set up to deliver its rated air and hold supply temperature; here the point is what the return is confessing about the air path.
What does a low delta-T mean?
A low delta-T at the cooling units almost always means bypass air: cold supply is short-circuiting back to the return without passing through a server. The return reads only a few degrees above supply because most of the air it carries never picked up any heat. This is the diagnosis to reach for first, because it is the most common cause and the cheapest to confirm.
Confirm it before you act. Read the supply temperature off the coil and the return into the unit, take the difference, and compare it to the rise the gear should be making at its real load. A unit delta-T well under the IT delta-T, with the load genuinely on, is bypass until proven otherwise. A return temperature that drops below the supply setpoint, so the unit returns air it never had to cool, is bypass in its most extreme form.
The reflex it triggers is the expensive mistake. A low delta-T makes the cooling look short, so the units ramp airflow and someone proposes more tonnage. Ramping airflow on a bypass problem just moves more cold air straight back to the return and drops the delta-T further. Chase a low delta-T as a leak in the air path, not as a cooling shortage, and the fix is usually sealing and balance, not equipment.
Bypass air: cold supply that goes home cold
Bypass air is cold supply that returns to the cooling unit without doing any work. It is the dominant driver of a low delta-T, and on a floor with open air paths it can be a large share of the total supply. Field measurements across many sites have put bypass in the range of 50 to 80 percent of supply air on floors with unsealed openings, which is air you paid to cool and then threw away.
It leaks through the usual paths. Unsealed cable cutouts in a raised floor let plenum air pour up behind the racks where no inlet can use it. Open U-spaces in a rack face let supply slip through to the exhaust side. A perforated tile set in a hot aisle dumps cold air straight into the return stream. And a floor simply over-supplied with more air than the gear can draw pushes the excess back to the units cold.
Every one of those drops the return temperature and the delta-T. The cure is the cheap air-side work: brush grommets on the cutouts, blanking panels in the open U-spaces, tiles only in the cold aisle, and airflow trimmed to what the gear actually draws. The data center airflow management and blanking panels guide is the full playbook for closing these paths. Close them and the delta-T comes up on its own, because the only way left for the supply air to reach the return is through a server.
Recirculation: hot exhaust looping back to the intake
Recirculation is the opposite fault. Instead of cold supply escaping to the return, hot exhaust loops back to the server intake, usually over the top of a rack, around the end of a row, or through an open U-space in the face. The inlet now breathes a blend of supply and exhaust, so it runs hotter than the supply temperature you set, and the gear at the top of the rack sees the worst of it.
The tell is a hot intake with a normal room. Recirculation does not lower the room average, so the floor reads fine while one rack alarms. It raises intake temperatures even where the supply is plenty cold, which is why managing to the room temperature hides it completely and managing to the intake catches it.
Recirculation and bypass often run on the same floor at once, which muddies the delta-T. Bypass pulls the return temperature down; heavy recirculation can pull it up, because the exhaust shortens its path back to the coil. The two can partly cancel in the unit reading while both are wrecking the inlets. This is why the per-rack intake map matters as much as the unit delta-T, and why the next two metrics exist to separate the two faults.
What is the return temperature index (RTI)?
The return temperature index, RTI, puts a number on the mixing the delta-T hints at. It is the unit delta-T divided by the IT delta-T, expressed as a percentage: RTI equals the return temperature minus the supply temperature, divided by the temperature rise across the IT equipment. A floor where the two rises match reads 100 percent, which is the target.
The deviation names the fault. An RTI below 100 percent points to bypass air, because the cooling airflow exceeds the rack airflow and cold supply short-circuits to the return without picking up the full rise. An RTI above 100 percent points to recirculation, because the rack airflow exceeds the cooling airflow and exhaust is being pulled back to the inlets. The metric formalizes exactly what reading the two delta-Ts against each other tells you, in one figure you can trend.
Treat the exact computation and any thresholds as something to confirm against the published method, since the metric uses airflow-weighted averages across the units and the racks, and definitions vary by source. RTI comes from Magnus Herrlin at ANCIS, the same source as the RCI, and that work is the usual reference. RTI is most useful as a before-and-after: run it before a sealing or containment project and again after, and the move toward 100 percent is the proof the air path got tighter.
RTI = [ (Treturn − Tsupply) / ΔTIT ] × 100%ΔT (°F) ≈ 3.1 × W / CFM (at sea level)- RTI = 100%
- Unit delta-T matches IT delta-T; the supply air does its work and returns fully heated
- RTI below 100%
- Bypass air; cooling airflow exceeds rack airflow and cold supply short-circuits to the return
- RTI above 100%
- Recirculation; rack airflow exceeds cooling airflow and exhaust is pulled back to the inlets
The rack cooling index (RCI): how well the intakes hold the range
Where RTI measures mixing, the rack cooling index, RCI, measures whether the intakes are actually in spec. It reports how well the rack intake temperatures stay inside the ASHRAE recommended range across the floor, and it comes in two halves. RCI HI tracks conformance at the hot end of the range, and RCI LO tracks the cold end. A floor where every intake sits inside the recommended band scores 100 percent on both.
The two metrics answer different questions and you want both. RTI tells you whether the air is doing its work; RCI tells you whether the gear is getting air inside the envelope it was built for. A floor can post a respectable RTI and still fail RCI HI if recirculation is cooking the top of a few racks, and a floor flooded with cold air can score RCI LO poorly while wasting energy. Read them together and you see efficiency and conformance at once.
RCI is a trademark of ANCIS and is computed from intake temperatures across the racks, so the score is only as good as the sensor coverage feeding it. Confirm the exact formula and the temperature limits against the published method and the ASHRAE class your equipment falls under, because the range that defines in-spec is the envelope, not a fixed pair of numbers.
Manage to the ASHRAE intake envelope, not the room
The temperature that matters is the one at the rack intake, and the standard that governs it is the ASHRAE TC 9.9 thermal guidelines. They give a recommended intake range, commonly cited near 18 to 27 C, or about 64 to 80 F, with wider allowable classes above and below for the equipment that can take it. Every delta-T and return-temperature decision serves keeping the intakes inside that envelope.
Managing to the room temperature is the error underneath most of the others. The room average is a blend that hides the spread, and the spread is the whole problem. A room reading a comfortable 22 C can have a top-of-rack intake at 32 C from recirculation and a flooded aisle at 16 C from bypass, both out of the efficient band, and the room number shows neither. Manage to the worst intake, not the average.
Confirm the actual numbers and the class against the current ASHRAE edition and the equipment's own ratings, because the envelope values move between cycles and the gear's listed limits can be tighter or wider than the general guideline. The recommended range is where you aim for efficiency and equipment life; the allowable range is how far you can go and for how long. Both are defined at the intake, which is why the intake is what you measure and what you control to.
Raising supply temperature, but only after the airflow is fixed
Raising the supply temperature is where the energy savings live, and a good delta-T is what makes it safe. A warmer supply means more economizer and free-cooling hours and a lower PUE for the same IT load, because the plant spends less of the year making cold. ASHRAE widened the recommended range specifically so operators could stop running the cold floors of a decade ago.
The order is the part people get backward. You cannot raise the supply on a floor that is mixing. If the intakes already run hot from recirculation, every degree you add to the supply lands on top of that and the hottest rack tips over. Fix the airflow first: close the bypass, kill the recirculation, get the delta-T up and the intake spread tight. Then the headroom to raise the supply appears, and you can see it instead of guessing.
Move the supply in steps and watch the worst intake, not the room average, because the average looks fine right until the one hot rack alarms. Raise it toward the top of the recommended band, confirm the intakes hold inside the envelope at full load, and let the units trim down. A tight delta-T turns a setpoint increase from a gamble into a measured move with margin you can defend.
Matching cooling airflow to IT airflow
The root cause of a bad delta-T is an airflow mismatch, so the fix is matching the cooling airflow to what the IT actually draws. The gear needs a volume of air set by its load and its design rise, commonly figured near 160 CFM per kW at about a 20 F rise for rack-mount servers, lower for blade gear, but confirm the number against the equipment. Deliver that and the gear makes its design delta-T.
Over-supply and under-supply fail in opposite directions, and the delta-T tells you which. Push more cold air than the racks can draw and the excess bypasses to the return cold, dropping the delta-T and the RTI below 100 percent. Push less than they draw and the racks pull the makeup out of the hot aisle, recirculating exhaust into the intakes and driving the RTI above 100 percent. The balance you want is the cold-aisle supply just meeting the rack demand, slightly positive against it, no more.
This is the discipline that separates a real balance from a paper one: balance the airflow to the load and to the intake, not to make the return look right. A floor tuned to satisfy a return average while a corner rack starves is a floor balanced to the wrong number. The CRAC and CRAH static pressure guide covers the plenum pressure and tile work that delivers the matched airflow rack by rack.
The fixes that raise the delta-T
Raising the delta-T is the same work as closing the air paths, done in order of cost. Blanking panels in the open U-spaces and seals on the rack rails and cabinet sides come first, because they stop recirculation through the rack for a few dollars per slot. Brush grommets on the floor cutouts and correct perforated-tile placement come next, closing bypass and holding plenum pressure. Right-sizing the airflow to the load, by pulling excess tiles and trimming fan speed, removes the over-supply that feeds bypass.
Containment is the capital step above the cheap fixes, and fan control to delta-T or pressure is the energy lever on top of all of it. The order holds because each cheap step reduces what the expensive one has to do. Seal and blank a floor and the hot spots can disappear before containment is ever scoped; the same sealing is also what makes containment pay.
Every item on that list moves the delta-T in the same direction, because they all force the supply air through a server before it can reach the return. The data center airflow management and blanking panels guide is the full sweep of these tactics. The point to carry here is that the delta-T is the scoreboard for the work: do the sealing and balancing, watch the unit delta-T climb toward the IT delta-T, and you have proof the air is finally doing its job.
How containment raises the delta-T directly
Containment raises the delta-T more directly than any other single move, because it stops the supply and exhaust from mixing at all. Doors on the ends of an aisle and a roof or ceiling seal the cold air apart from the hot, so the supply has nowhere to go but through the racks. The return comes back fully heated, the unit delta-T climbs to match the IT delta-T, and the RTI moves toward 100 percent.
The mechanism is simple. Without containment, cold supply and hot exhaust share the room and blend at every gap, so the return is always a dilution of the two and the delta-T is always depressed. Separate them and the dilution stops. A contained floor routinely runs a return temperature high enough to let the supply rise into the top of the recommended band, which is where the economizer hours and the energy savings come from. Hot-aisle or cold-aisle containment is commonly credited with cutting cooling energy on the order of 15 to 20 percent, though the real figure depends on the floor and the climate.
Containment without the cheap fixes underneath it is wasted money, because it raises the pressure across any leak you left open and makes that leak worse. Blank and seal first, then contain. The hot-aisle versus cold-aisle choice, the differential-pressure setpoint, and the acceptance test belong to the containment scope; the delta-T point is that a sealed aisle is the most reliable way to get the rise where the design wanted it.
Controlling the fans to delta-T, and the cube law
Control the fans to the delta-T or the plenum differential pressure, not to a fixed speed, and the energy follows the cube law down. Modern CRAC and CRAH units run EC plug fans or VFD-driven fans, and fan power rises with the cube of speed. Trim a fan to 80 percent and you get about 80 percent of the airflow for roughly half the power. On a tight floor that trim is the largest single fan-energy lever there is.
The control logic is the point. If the units hold a return-temperature or inlet-temperature target by modulating fan speed, the airflow tracks the load automatically: as racks draw more, the fans ramp to feed them; as the load eases, they back off and the cube law banks the savings. Fixed-speed fans cannot do this, so they over-supply at light load, which is itself a bypass generator and a depressed delta-T.
The catch is the fan curve. Trim too far and the plenum pressure collapses at the far tiles, and the same cube law that saved energy now starves a row. Set the minimum fan speed to hold the worst tile, not the average, and pick the control reference to suit the floor. Return-air control reacts to an average and misses the hot rack; supply-air or cold-aisle control protects what you actually care about. The CRAC and CRAH static pressure guide walks the affinity laws and the control-strategy choice in full.
Measure per rack and per unit
You manage what you measure, and on the air side that means the delta-T per cooling unit and the intake temperature per rack. The unit delta-T comes off each unit's controller as return minus supply. The IT delta-T comes from intake and exhaust readings at the racks. Trend both and you can see bypass and recirculation develop before they become a hot-spot call.
Take the readings under real load. Airflow faults only exist when the gear is moving air, so a floor measured at idle tells you almost nothing about how it behaves at full draw. Read the units and the racks as close to the same conditions as you can, because the two delta-Ts only mean something when they are compared against each other on the same load.
A DCIM platform or rack-mounted environmental sensors give the permanent picture, and a thermometer or thermal camera covers a walk-through. Whatever the sensing, the discipline is recording it and tying it to a baseline, so the warm aisle six months out is a check against a known-good state instead of a fresh investigation. FieldOS is built for this kind of field record: capture the per-unit delta-T and the per-rack intake map, photograph the readings against the rack, and keep the as-found and as-corrected numbers attached to the location so the next visit reads against them instead of starting over.
Where to put the sensors: intake and return
Put the intake sensors at the rack face, and put them at three heights: top, middle, and bottom. The top of the rack is where recirculation shows up first, because exhaust rolls over the top of the cabinet and lands on the upper intakes, so a single sensor at mid-height misses the hottest reading on the rack. The bottom sees the coldest supply. The spread between top and bottom is the recirculation signal the average would hide.
ASHRAE measurement guidance treats the rack intake as the point of compliance and uses multiple-height sensing for exactly this reason. Confirm the placement and the number of points against the current guidance and your monitoring scheme, since the count and the heights vary with rack height and how the floor is instrumented. The principle does not change: measure where the gear breathes, not where it is convenient.
On the cooling side, read the return into the unit and the supply off the coil, and put the return sensor where it sees the real return stream, not a pocket of mixed room air. A return sensor sitting in a dead zone reports a number that flatters the unit and hides a low delta-T. Place it in the return path, trend it against the intake map, and the two together tell you what the air is doing across the whole hall.
The capacity you free by raising the delta-T
Raising the delta-T frees cooling capacity that was already on the floor, because a cooling unit's real output rises with its return temperature. A unit starved of return heat delivers far less than its nameplate; feed it a warmer return and it earns its rating. As a published example of the effect, a 20-ton CRAC rated near 84 kW at a 75 F return can deliver on the order of 137 kW at a 90 F return, per the manufacturer's capacity curve. The hardware did not change. The return temperature did.
That recovered capacity is real money. Fewer units can carry the same load, which means units can be turned off for redundancy or service instead of all running flat out, and the next cooling purchase can be deferred or canceled. The capacity people scope a new unit to buy is frequently sitting stranded inside the units they already own, locked up by a low delta-T.
This is the argument that justifies the sealing and balancing work to a budget. The cheap air-side fixes raise the delta-T, the higher delta-T recovers rated capacity, and the recovered capacity defers capital. Put the before-and-after delta-T and the freed tonnage in the record, because a stranded-capacity recovery you can show on paper is what turns the next hot-spot call away from a purchase order and toward a balance.
| Return temperature at the unit | What it signals | Effect on capacity |
|---|---|---|
| Low return, low delta-T (a few degrees) | Heavy bypass; cold supply short-circuiting | Output far below nameplate; capacity stranded |
| Return rising toward design | Air path tightening; bypass closing | Output climbing toward the rating |
| Return at design, delta-T matches IT | Air doing its work; RTI near 100 percent | Unit delivers its rated capacity |
| Example: 20-ton CRAC, 75 F return | Typical low-return operation | Near 84 kW (per manufacturer curve) |
| Example: same unit, 90 F return | Warm return from a tight floor | On the order of 137 kW (per manufacturer curve) |
Field example: a low delta-T that read as a cooling shortage
A hall flagged a cooling shortage. Several CRAH units were pinned near full fan speed, a row was running hot at the top of the racks, and the operations team had a quote in hand for an added unit. The unit delta-T told a different story: the units were returning air at about 7 F over supply while the gear in that row was making close to a 20 F rise under load. The return was cold because most of the air never touched a server.
The walk-down found the leaks. A run of cable cutouts behind the hot row was wide open, pouring plenum air straight up into the return path, and two perforated tiles had been added in the hot aisle to chase the warm top-of-rack, which only fed more cold supply into the exhaust. The intake map showed the classic split: cold at the bottom of the racks, hot at the top, with the room average reading fine.
The fix was a case of brush grommets, blanking panels in the open U-spaces, pulling the two misplaced tiles, and trimming fan speed once the bypass was closed. The unit delta-T came up to about 18 F over the next shift, the RTI moved toward 100 percent, the hot-rack intakes dropped inside the envelope, and the added unit was canceled. The capacity had been there the whole time, stranded by a low delta-T.
The lesson is the one this guide opens on. The low delta-T was the early warning the team had been reading as good news. Once it was read as a leak gauge, it pointed straight at the air paths, and the cure cost a fraction of the tonnage that was nearly purchased to paper over it.
| Reading | As found | After sealing and balance |
|---|---|---|
| Unit delta-T (return minus supply) | about 7 F | about 18 F |
| IT delta-T across the row | about 20 F | about 20 F |
| RTI (unit delta-T / IT delta-T) | roughly 35 percent (bypass) | near 100 percent |
| Top-of-rack intake | over the ASHRAE band | inside the recommended band |
| Fan speed | pinned near full | trimmed on delta-T |
| Units added | one quoted | none |
What to document
A delta-T problem you cannot show is a delta-T problem the next person re-discovers from scratch. The record is the as-found and as-corrected state of the air side, and it is what every future hot-spot call gets checked against. Capture it per unit and per rack, under real load, with the date and the conditions, so a reading six months out means something.
Record the per-unit return and supply temperatures and the delta-T, the IT delta-T across the loaded rows, the RTI and RCI if you compute them, the rack intake temperatures top to bottom, and the supply setpoint with the envelope it was set against. When you seal, balance, or raise a setpoint, write the before and after, because a change you cannot show is a change the next person will not trust. Hedge the targets in the record to the ASHRAE envelope, the project basis of design, and the equipment ratings, since those control the numbers, not a rule of thumb.
| Field to record | Why it matters |
|---|---|
| Per-unit return, supply, and delta-T | The unit delta-T is the first bypass signal |
| IT delta-T across loaded rows | The match against the unit delta-T is the diagnosis |
| RTI and RCI, if computed | Separates bypass from recirculation, and tracks conformance |
| Rack intake temps, top to bottom | The spread is the problem the room average hides |
| Supply setpoint and target envelope | Ties the warm-running decision to ASHRAE and the gear |
| Before and after on each change | Proves the delta-T moved and the capacity was freed |
| Load and conditions at the reading | Airflow faults only show under real load |
Common mistakes
- Reading a low delta-T as good news instead of as the bypass signal it usually is.
- Over-supplying airflow so cold supply short-circuits to the return and depresses the delta-T.
- Running with no hot-aisle or cold-aisle containment, so supply and exhaust mix and dilute the return.
- Managing to the room temperature instead of the rack intake, which hides the spread entirely.
- Missing recirculation hot spots because the top-of-rack intake is never measured.
- Ramping fan speed or scoping more tonnage to answer a low delta-T caused by a leak.
- Raising the supply setpoint before the airflow is fixed, so the hottest rack tips over.
- Taking no per-rack or per-unit measurement, so the delta-T problem is invisible until a rack alarms.
Field checklist
Want this checklist to run itself on every job — with photo proof and a signed record crews can hand the customer? That's FieldOS.
Standards and references
The thermal target every delta-T decision serves comes from ASHRAE Technical Committee 9.9 and its Thermal Guidelines for Data Processing Environments, which set the recommended and allowable intake envelopes and the equipment classes. The recommended intake range is commonly cited near 18 to 27 C, with wider allowable classes, but the values and the classes move between editions, so confirm the current numbers against the published guideline and the equipment manufacturer. Manage to the intake the guideline defines, not the room.
The air-management metrics come from their authors, with RTI and RCI from Magnus Herrlin at ANCIS and The Green Grid publishing related air-management guidance. The return temperature index, RTI, is the unit delta-T over the IT delta-T, with 100 percent the target, below pointing to bypass and above to recirculation. The rack cooling index, RCI, a trademark of ANCIS, measures intake conformance at the hot and cold ends of the recommended range. Treat the exact formulas, the weighting, and any thresholds as values to verify against the published method, because the definitions vary by source and by how the floor is instrumented.
The supporting bodies sit underneath. ASHRAE Standard 90.4 bounds the mechanical energy overhead for data centers; AABC and NEBB own the test-and-balance procedures a witnessed air balance follows; the Uptime Institute publishes the operational practice and the redundancy demonstrations; TIA-942 covers the broader infrastructure. Underneath all of it sit the project basis of design and the equipment ratings, which set the real delta-T target, the supply temperature, and the airflow per kW. Cite the body that owns the point, hedge to the softer claim when a value is project-dependent, and let the equipment ratings override a rule of thumb when they are stricter. The two things that do not bend: raise and match the delta-T, and measure it per rack and per unit.
Units, terms, and conversions
Delta-T work borrows vocabulary from HVAC, from the IT side, and from the air-management metrics, and the same idea reads differently across a controls screen, a DCIM dashboard, and a vendor sheet.
Delta-T is a temperature difference in degrees F or C; a 20 F rise is about an 11 C rise. Airflow is CFM in the field, or liters per second and cubic meters per hour in metric sources, commonly near 160 CFM per kW at a 20 F rise. Return and supply temperatures are read at the cooling unit; intake and exhaust are read at the rack. The ASHRAE ranges are usually stated in C. RTI and RCI are percentages, with 100 percent the target for each.
- Delta-T
- The temperature rise of the air across the IT equipment, or return minus supply across the cooling unit
- Return temperature index (RTI)
- Unit delta-T divided by IT delta-T, as a percent; below 100 is bypass, above 100 is recirculation
- Bypass air
- Cold supply that returns to the cooling unit without passing through any equipment, lowering the delta-T
- Recirculation
- Hot exhaust pulled back into the rack intakes, raising intake temperature above the supply
- Rack cooling index (RCI)
- How well rack intake temperatures stay inside the ASHRAE recommended range, at the hot and cold ends
- ASHRAE recommended and allowable
- The TC 9.9 intake envelopes; recommended is the efficiency aim, allowable is the outer limit, both at the intake
FAQ
What is delta-T in a data center?
Delta-T in a data center is the temperature rise of the air as it passes through the IT equipment, inlet to exhaust. The same rise should show up at the cooling unit as return minus supply. When the two match the load, the air is doing its work and the floor is running efficiently.
What does a low delta-T mean at the cooling units?
A low delta-T at the cooling units usually means bypass air: cold supply is short-circuiting back to the return without passing through a server. The return reads only a few degrees above supply, so the unit moves a huge volume of air for little heat removal, wasting fan energy and stranding capacity. Confirm it before adding cooling.
What is the return temperature index (RTI)?
RTI is the unit delta-T divided by the IT delta-T, as a percent: return minus supply, over the rise across the IT equipment. The target is 100 percent. Below 100 percent points to bypass air; above 100 percent points to recirculation. Confirm the exact computation and thresholds against the published method, since definitions vary.
What is bypass air in a data center?
Bypass air is cold supply that returns to the cooling unit without passing through any equipment. It leaks through unsealed floor cutouts, open rack U-spaces, and tiles placed in hot aisles. It lowers the return temperature and the delta-T, and on floors with open paths it can be 50 to 80 percent of supply air.
How do I raise the delta-T in my data hall?
Close the air paths that let supply skip the servers. Fit blanking panels and rack seals to stop recirculation, brush grommets on floor cutouts to stop bypass, and place perforated tiles only in the cold aisle. Then match the airflow to the load and contain the aisle. The delta-T climbs toward the IT rise as the leaks close.
Should I manage to the room temperature or the rack intake?
Manage to the rack intake, not the room. The room average is a blend that hides the spread, and the spread is the problem. A comfortable room can have a top-of-rack intake well over the ASHRAE band from recirculation and a flooded aisle below it from bypass. Measure at the intake, top to bottom, and control to the worst one.
How much cooling capacity does raising the delta-T free up?
A cooling unit's output rises with its return temperature, so a warmer return earns more of its rating. As a published example, a 20-ton CRAC near 84 kW at a 75 F return can deliver on the order of 137 kW at a 90 F return. That recovered capacity lets you turn units off or defer a purchase.
What is the rack cooling index (RCI)?
RCI measures how well rack intake temperatures stay inside the ASHRAE recommended range, in two halves: RCI HI at the hot end and RCI LO at the cold end, with 100 percent meaning all intakes in spec. Where RTI tracks mixing, RCI tracks conformance. Confirm the formula and limits against the published method and your ASHRAE class.
Can I raise the supply temperature to save energy?
Yes, but only after the airflow is fixed. A warmer supply within the ASHRAE envelope buys economizer hours and a lower PUE. On a mixing floor, every degree you add lands on intakes that already run hot and the worst rack tips over. Tighten the delta-T first, then raise the supply in steps while watching the worst intake.
Why does my return air read colder than the supply setpoint?
That is bypass in its most extreme form: so much cold supply is short-circuiting to the return that the air comes back below the setpoint, and the unit returns air it never had to cool. It signals a badly over-supplied or unsealed floor. Close the cutouts and open U-spaces and trim the over-supply, and the return temperature recovers.
People also ask
Codes cited in this guide
This guide is written and reviewed against the published standards below. Always confirm the current adopted edition with the authority having jurisdiction.