From Server Supply and Return to Facility Water: Deriving the Primary-Loop Temperature Targets

Every direct-to-chip project reaches the same hand-off meeting. The server team arrives with two numbers — coolant in at 32 °C, out at 42 °C — and asks the facility team what chilled water it needs. The facility team arrives with two different numbers, usually 7 and 12, and asks what the servers want. Both sides are holding genuine specifications, and neither can be derived from the other without the middle of the system: the coolant distribution unit and the temperature it spends between the two loops.

That middle is not complicated. It is three subtractions on the supply side and one heat balance on the return side. Written down in order, the primary-side — facility — supply and return targets fall out on their own, and you also learn which ASHRAE water class you are actually building to, how many free-cooling hours you just bought or surrendered, and whether your heat-rejection plant can make that water at all on a design day. This reference works the chain end to end, quantifies what each kelvin costs, and lists the places where the arithmetic reliably goes wrong. For the hydraulics on the same hand-off, see our CDU pipe sizing guide and the CDU maintenance schedule.

Name the loops before you do any arithmetic

Almost every argument about these temperatures is a vocabulary argument in disguise, because three loops have to be told apart and two of them have colliding names.

The technology cooling system (TCS) is the loop the CDU pumps: coolant from the CDU, through the rack manifolds, into the cold plates, and back. The facility water system (FWS) is the building or campus loop that carries the heat away to chillers, cooling towers, dry coolers or a heat-reuse connection. In ASHRAE’s usage — and this is the convention worth locking — the primary side of the CDU heat exchanger is the facility water side and the secondary side is the technology cooling side [1]. Chinese practice maps the same words the same way: the plant loop is 一次侧 and the server loop is 二次侧.

Table 1 — The four temperatures, their two vocabularies, and which standard owns each. W classes describe what the plant can deliver; S classes describe what the silicon will accept [1][3][4].

Two details in that table do more work than they look like. First, the S-class minimum is a dew point rule, not an efficiency choice: ASHRAE holds TCS supply at least 2 °C above room dew point to keep condensate off energised electronics, and the CDU is the component expected to enforce it [3][6]. Second, the W-class number is a ceiling, not a setpoint — a W32 facility is one whose supply never exceeds 32 °C, and a design target anywhere below that ceiling lives inside the class. That is what makes the class you need a derived quantity rather than a preference [1][5].

The supply side is three subtractions

Start from the only number the silicon vendor actually gives you: the maximum coolant temperature at the server inlet. Everything else on the supply side is what the path from the plant to that inlet takes away.

Step 1 — the line pickup. The CDU controls its own supply temperature; the server feels it at the rack manifold, after however much heat the supply hose and manifold collect or shed on the way. Deduct 1 K as a design allowance for that pickup plus control tolerance and sensor drift. The honest sign of this term depends on the design: in a cold-water loop the coolant is colder than the room and the line warms it, so the deduction is real. In a warm-water loop the coolant is warmer than the hall and the line cools it, returning a little margin. Deducting 1 K everywhere is the conservative choice, and it also covers the meter you will wish you had.

Step 2 — the approach. This is the whole conversation. The approach temperature difference is the gap between TCS supply and FWS supply across the CDU’s plate heat exchanger — the same gap, at the same end of the exchanger, in every credible definition of the term [6][7][8]. It is the only number on the supply side that is genuinely negotiable, which is why this article spends a full section on it. OCP rates liquid-to-liquid CDU capacity at an approach of 5 °C [6]; well-sized machines land between 2 and 5 K, with plate exchangers in CDU service typically built for 2–4 K [9][10].

Step 3 — hand it to the plant. The facility supply target is the leaving-water setpoint of whatever makes the water, plus whatever the plant’s own distribution piping picks up between the plant room and the CDU. If the plant is 200 m away outdoors, that last term is real and it is a summer number, not an annual average.

Run the chain in reverse and you get the table that settles the hand-off meeting — the server inlet a given facility class can support, or equivalently the class a given server inlet obliges you to build.

Table 2 — The master mapping, at a 1 K line pickup and a 4 K CDU approach. Read it both ways: a server limited to a 32 °C inlet obliges a W27-class plant, and a server that tolerates 45 °C at the inlet unlocks W40 and year-round dry cooling. The W45 row is why vendors pushing warm-water designs quote 50 °C-class inlets. W-class ceilings per ASHRAE [1][5]; the deductions are yours.

One caution on how this table gets used. No OEM document actually names an ASHRAE water class for its rack; what vendors publish is an inlet temperature limit, and the class is your derivation from it [11]. Treat a datasheet that claims “W45 compatible” as a summary of an inlet limit, and re-derive it with your own line pickup and your own approach — the two terms the vendor did not use.

The return side is an output, not a setting

The supply target comes from a subtraction chain. The return target does not come from anywhere — it is what the heat balance leaves behind, and pretending otherwise is the second most common error in these calculations.

Steady state, the same power crosses the plate exchanger on both sides:

Two consequences follow, and both bite people who copy ΔT across the exchanger. First, ΔT is not conserved: the primary ΔT is the secondary ΔT scaled by the flow ratio and by the specific-heat ratio, and on a PG25-to-water pair the cp ratio alone is 3.95/4.186 ≈ 0.94 [12]. Second, the primary ΔT is a design choice made with the flow ratio, not a property of the exchanger.

Take the same 900 kW hall, TCS supply 31 °C, TCS return 42 °C, facility supply 27 °C, and vary only how much primary water you run:

Table 3 — The primary return is whatever the flow ratio says it is. Less primary water buys a hotter return; more buys pipe and pump. All five rows reject exactly the same 900 kW.

Which direction is right? A hotter return is usually the better trade: it is the number a cooling tower, a dry cooler or a district-heating offtaker actually cares about, it lets the rejection plant run a wider ΔT on less flow, and it is what makes water-side free cooling and heat reuse productive. The costs run the other way — a tight primary flow raises the primary-side pressure drop the plant pumps against, and it moves the whole plant closer to the low-ΔT cliff if anything in the building dilutes the return. Where primary and secondary sit on either side of a decoupler, keep the common-pipe velocity near 0.6–1.2 m/s so the two loops stop arguing with each other [13].

A worked example, end to end

Here is the full chain for a 1 MW hall, 90 % of the load on liquid — a realistic liquid fraction, since even a fully liquid-cooled rack leaves power shelves, NICs and drives on room air; one published NVL72 breakdown puts 115 kW of a 132 kW rack on liquid and 17 kW on air [11].

Worked example — one hall, six numbers in, four out. The 1.35 L/min per kW that falls out sits right under the 1.5 L/min per kW at a 10 K rise that OCP guidance uses as a planning basis, the difference being the line pickup [6][14].

Two footnotes make this example honest. Flow is often set by the cold plate rather than by the heat balance: a published NVL72 briefing pairs a 25 °C inlet with a 45 °C outlet at roughly 120 L/min per rack, which implies a design ΔT nearer 14 K once the cold-plate pressure-drop requirement has had its say — and no OEM publishes a flow table specific to GB200, so the figure you use has to come from a quote [15][16]. And the TCS supply of 31 °C must clear the dew-point floor with room to spare, which in most halls it does easily; in a humid hall running a cold loop it is the binding constraint, not the silicon.

What one kelvin of approach costs, and what it buys

The approach is the only supply-side number you can trade, so price it honestly in both directions.

What it costs to shrink. Halving the approach roughly doubles the exchanger. One CDU manufacturer’s published selection, at 1,000 kW with a 10 K ΔT on both sides, needs 71.8 m² of plate area at a 4 K approach and 143.6 m² at 2 K — the same duty, twice the metal, twice the footprint and a wider primary pressure drop [12]. Industry guidance is blunt about the floor: 2–5 K is the workable range for a CDU plate exchanger, and forcing 1 K buys area, connection diameter and pressure-drop sensitivity faster than it buys capacity [10]. A draft Chinese group standard for CDU plate exchangers sets the approach requirement at ≤3 °C and, usefully, caps the design fouling resistance at 0.0002 m²·K/W for water and 0.0003 for glycol — the two numbers that decide whether your 3 K approach is still 3 K in year four [17].

What it costs to waste. Every kelvin of approach is a kelvin of facility supply you gave back, and a kelvin of supply temperature is the cheapest efficiency there is. LBNL’s rule of thumb is 1–1.5 % chiller efficiency per °F of chilled-water supply reset — roughly 2–3 % per kelvin — and its worked example takes a constant-speed machine from 0.63 to 0.40 kW/ton across a 6.7 → 15.6 °C reset [18]. Operators report the same slope from the other direction: 2–2.5 % per kelvin on mainstream centrifugal plant, with water-side free-cooling hours expanding by roughly 1,500 hours between a 12 °C and an 18 °C supply in a subtropical climate [19].

And what it costs to mis-read. The approach is a rated condition, not a constant. Comparing CDU capacities quoted at different approaches is described by people who see a lot of datasheets as the single most common selection error — a unit rated 300 kW at a 7 K approach and a unit rated 300 kW at 3 K are not the same machine, and only one of them will hold your supply target with warm facility water [7][8]. In operation, fouling and chemistry drift widen the approach over time, which is why approach and exchanger Δp belong on the trend list next to flow.

The check everyone skips: can the plant make that water?

The supply target is now a number. The next question is whether anything in your climate can produce it on a design day, and this is where the temperature budget fails silently — the rejection plant gets blamed for a mismatch that was decided three subtractions earlier [20].

Table 4 — What each rejection machine can approach, and the air temperature it approaches it against [21]. Tightening any of these is a surface-area purchase: halving a dry-cooler approach roughly doubles the coil [22].

Now test the worked example’s 27 °C target against a 32 °C dry-bulb / 24 °C wet-bulb design day:

Table 5 — The design-day verdict for a 27 °C facility supply. Note how close the tower comes: one kelvin of tower approach, or one kelvin of CDU approach, flips the verdict. That is the lever the whole hand-off is really negotiating.

This is also why the same server hall can be a chiller plant in one climate and a dry-cooled plant in another with no change to the silicon. Published sizing guidance for warm-water direct-to-chip lands on a 3–5 K CDU approach, which yields a 30–37 °C facility supply and lets a dry cooler ride the local 99 %-design wet bulb — in London, roughly a 20 °C wet bulb, so a 7 K dry approach still makes 35 °C supply on the hottest day [23]. The moment the server inlet limit drops, every one of those hours is repriced.

The floor: dew point, not efficiency, ends the cold end

Nothing above says why you would not simply run the facility water colder and bank the COP. The answer is condensation. ASHRAE’s S-class floor holds TCS supply at least 2 °C above room dew point, and the CDU is the component charged with measuring the room and lifting the setpoint to respect it [3][6]. The W-class floor of 2 °C exists for the same reason at the plant end [1].

That floor binds hardest exactly where retrofit projects most want cold water. A hall holding a 15 °C dew point cannot run TCS supply below 17 °C — which lands the design on the W17 boundary and drags a chiller back into a system that was otherwise dry-coolable. It is one more reason the supply-side subtractions should be written down before equipment is selected: the cold end of the range is capped by the air in the room, and the warm end is capped by the silicon.

Ten traps, in the order they are met

1. Using the server inlet limit as the CDU setpoint. They differ by the line pickup. Set the CDU 1 K below the limit, or the rack discovers the difference on a hot afternoon.
2. Copying ΔT across the exchanger. The primary ΔT is the secondary ΔT times a flow ratio times a cp ratio. Equal mass flow on PG25-to-water already loses 6 % to cp alone.
3. Comparing CDU capacities at different approaches. OCP’s rating point is 5 K; a vendor quoting at 3 K is selling a bigger exchanger, and the kW are not comparable [6][7].
4. Sizing the return at design flow and being surprised at part load. The CDU’s modulating valve throttles primary flow as load falls, the primary ΔT stretches, and the return runs hotter than the design number. Design the rejection plant for the part-load return too — this is the same part-load family of problems as running a liquid hall chronically below design load.
5. Treating 7/12 as a law of nature. It is the AHRI 550/590 standard rating condition, 6.7 °C leaving and 12.2 °C entering, chosen so that published efficiencies are comparable — not the temperature your loop owes the silicon [24].
6. Forgetting the liquid fraction. A 132 kW rack does not put 132 kW into the loop; roughly 115 kW does, and the rest heats the hall [11]. Size both systems for their own share.
7. Forgetting the rejection-side approach. The facility supply target and the design wet bulb are separated by the tower or dry-cooler approach, and that gap is where the trim chiller comes from [20][21].
8. Ignoring the dew-point floor. The cold end is capped by room humidity, and the cap is enforced by the CDU, not by the operator’s preference [3][6].
9. Assuming the approach holds for the life of the plant. It degrades with fouling, which is why the design fouling resistance belongs in the specification and the measured approach belongs on the trend list [17].
10. Letting W and S blur. W describes the plant, S describes the silicon, and the approach is the only thing that connects them. One published example makes the linkage explicit: facility water at W35 with a 5 °C approach yields TCS supply at S40 [8].

Commissioning: what to measure before you sign off

The derivation above is arithmetic, and arithmetic survives contact with reality only if you verify three things.

The heat balance closes. In steady state, primary-side power must equal secondary-side power plus pump heat, within a few percent. A balance that will not close points at a flow meter, a sensor in the wrong place, a bypass path, entrained air or simply an unsteady operating point — and it is the single cheapest commissioning check in the building [6].

The approach is measured at several loads, not one. Record TCS supply, FWS supply and the gap at 25, 50, 75 and 100 % of design load, and file the numbers. The gap at design load is the number the datasheet promised; the trend across the year is the number that catches fouling [17].

The dew-point interlock acts. Raise the room dew point on test and confirm the CDU lifts the TCS supply setpoint and alarms — not just logs. This is the control loop that prevents the one failure mode in this whole chain that damages hardware rather than efficiency [6].

Help is arriving on the comparability front. ASHRAE published Addendum b to Standard 127-2020 in February 2026, establishing a method of test for rating liquid-to-liquid CDUs — until now there was no standardised industry rating for most L2L units and procurement leaned on factory acceptance tests [25]. ASHRAE’s 2025 Technical Alert set out the isolation of TCS from FWS as the core reliability function, and OCP and ASHRAE formed a formal alliance in October 2025 to harmonise the two specifications. Within a procurement cycle or two, the approach will be a rated, comparable, contractable number rather than a datasheet footnote.

The worksheet

Everything above reduces to five lines. Fill them in at the hand-off meeting and the argument ends there.

Two of the five lines are given, one is a purchase, one is a balance and one is a weather question. Nothing about the hand-off is mysterious once it is written in that order — and the facility side stops being asked to match numbers it was never shown how to derive.

Planning a liquid-cooled hall and want the temperature budget, the hydraulics and the rejection plant sized against your actual server specification? Send us the server numbers and the site and we will work the chain with you.

References

1. ASHRAE Dallas Chapter, "The ASHRAE Thermal Guidelines for Data Centers: Past, Present, and Future" (April 2025) — FWS thermal guideline table: W17 17 °C, W27 27 °C, W32 32 °C, W40 40 °C, W45 45 °C, W+ above 45 °C, minimum water temperature 2 °C for all classes; TCS fluid classes S30/S35/S40/S45/S50 with minimum TCS temperature 2 °C above room dew point. dallas-ashrae.org (PDF)
2. ASHRAE TC 9.9 committee page — the rewritten Liquid Cooling Guidelines introduce "a new set of environmental classes (S Classes) for the supply temperature of the Technology Cooling System (TCS) coolant", plus expanded treatment of TCS/FWS fluid quality separation. tpc.ashrae.org
3. ASHRAE TC 9.9 members on the ASHRAE Journal Podcast (June 2025) — the S-class range "which was S-30 to S-50 has expanded to S-20 to S-50", with S-20 corresponding to a 20 °C inlet to the IT equipment, and the rationale that facility water classes alone no longer describe what the IT equipment accepts. ashrae.org via Castro.fm transcript
4. Compute Law Blog, "Cooling system construction contracts for AI data centers" — S classes now range from S20 to S50; large CDU approach temperatures leave ITE inlet temperatures 3–15 °C or more above facility water supply; ASHRAE Addendum b to Standard 127-2020 (published 27 February 2026) establishing a method of test for rating CDUs; ASHRAE Technical Alert 2025 on TCS/FWS isolation; the OCP–ASHRAE alliance of October 2025. computelaw.blog
5. OCP, "Data Center Liquid Distribution Guidance & Reference Designs" white paper — ASHRAE TC 9.9 liquid-cooling class table (W17/W27/W32/W40/W45/W+) with facility water supply temperatures and typical primary/supplemental equipment, minimum 2 °C, and the design note that cold-plate and immersion loops operating at W32 and above enable free cooling without condensation concerns. ocp-all.groups.io (PDF)
6. Harris Thermal, "AI data center liquid cooling: where heat exchangers fit in the cooling infrastructure" (2026) — in ASHRAE usage the primary side of the CDU exchanger is the facility water side and the secondary is the technology cooling side; the CDU "is responsible for monitoring the ambient dew point and elevating the secondary water loop (TCS) supply temperature to at least 3.6 °F (2 °C) above the room dew point"; Uptime observes facility water temperatures converging around 32 °C; FWS/TCS water-quality domains roughly an order of magnitude apart. harristhermal.com
7. Keentel Engineering, "Data Center Coolant Distribution Units (CDU) Guide" — approach temperature difference defined as TCS supply minus FWS supply; OCP’s liquid-to-liquid CDU test methodology rates capacity at a 5 °C ATD; ASHRAE W-class table with typical rejection equipment; the heat-balance closure check as a commissioning and troubleshooting tool; OCP flow guidance of about 1.5 L/min per kW at a 10 K rise with PG25 at 24.5–27.5 % by volume. keentelengineering.com
8. Aulank, "What Is a Coolant Distribution Unit (CDU)?" — approach defined as the difference between secondary supply and primary supply; "comparing units rated at different approach temperatures is the most common datasheet error", with a 300 kW at 7 °C approach example; primary/secondary fluid separation and its rationale. aulankpump.com
9. JetCool, "Coolant Distribution Units for AI" — the W/S linkage worked explicitly: facility water at W35 with a 5 °C CDU approach yields TCS supply at S40; ASHRAE TC 9.9 Technical Alert on TCS/FWS isolation; CapEx caution against oversized CDUs. jetcool.com
10. STULZ, "Liquid Cooling: When performance meets sustainability" white paper (2025) — CDU plate-exchanger selection cases at 1,000 kW: FWS 22/32 °C, TCS 26/36 °C at a 4 K approach requires 71.8 m² of heat-transfer area, and the same duty at a 2 K approach requires 143.6 m²; PG25 specific heat 3.95 kJ/kg·K against 4.183 for water; brownfield 20/30 °C FWS practice; TCS requirements of 35/25 and 36/26 °C in greenfield projects. stulzasia.sg (PDF)
11. SecondWatt, "GB200 NVL72" — the loop takes up to 45 °C inlet, so a dry cooler suffices in most climates; 115 kW to liquid and 17 kW to air on a 132 kW rack; "no OEM document names an ASHRAE water class"; in-rack 250 kW, in-row 1.3 MW and liquid-to-air sidecar 180/240 kW CDU options. secondwatt.com
12. Ekin Industrial, "Data Center Liquid Cooling" — a 2–5 K approach is applicable to a CDU plate heat exchanger and forcing 1 K significantly increases plate area, connection diameter and pressure-drop sensitivity; technology-loop ΔT typically 5–12 K; Q = ṅ·cp·ΔT worked flows (1,000 kW at 8 K = 29.9 kg/s = 108 m³/h versus 215 m³/h at 4 K); closed-loop operating pressure 2–6 bar; control target 2–3 K above dew point. en.ekinendustriyel.com
13. Ingener HVAC Encyclopedia, "Chilled Water Systems for Cooling Applications" — Q = ṅ·cₖ·ΔT in standard units; design ΔT conventions from 10 °F traditional to 16–20 °F high-ΔT; pump power proportional to the cube of flow (a 20 % flow increase costs 73 % more pumping energy); primary–secondary common-pipe velocity kept at 0.6–1.2 m/s and the decoupler sized to prevent loop interaction. ingener.by
14. Resistance Zero, "CDU Install, Inspection & Maintenance Checklist" — secondary supply band ~17–45 °C against ASHRAE W17–W45; secondary ΔT ~8–12 °C; OCP flow guidance 1.0–1.5 LPM/kW at a 10 °C rise; L2L approach (ATD) 3–5 °C with higher ATD indicating fouling; supply held 2–3 °C above room dew point as a critical interlock. resistancezero.com
15. Enzu, "GB200 NVL72 Price (2026)" — NVIDIA’s 2024 briefing gave about 25 °C coolant in, about 45 °C out and about 2 L/s (roughly 120 L/min) per rack; "we found no OEM flow table specific to GB200, and OEM tables for the next generation show flow rising steeply with warmer inlet water, so ask your OEM for its exact figure"; 132 kW total with 125–135 kW operating; peak/transient up to 192 kW. enzu.com
16. GPUSmith, "NVIDIA GB200 NVL72" datasheet — 72 B200 GPUs and 36 Grace CPUs in one NVLink domain; 125–135 kW operating rack power; 100 % direct-to-chip with zero rack fans; facility coolant loop reported as ASHRAE W-series with supply 18–25 °C and return up to 45 °C depending on CDU configuration; in-rack CDU up to 250 kW or 1.3 MW in-row. gpusmith.com
17. China Group Standard (drafting explanation), "Data Center Liquid-Cooling CDU Plate Heat Exchanger" — approach requirement set at not more than 3 °C with the rationale that a smaller approach improves PUE and demands an overall heat-transfer coefficient of at least 3,000 W/(m²·K); fouling factor capped at ≤0.0002 m²·K/W for water and ≤0.0003 for glycol, based on performance decay after 500 hours of continuous CDU operation; burst pressure not less than five times design pressure. ttbz.org.cn (PDF, Chinese)
18. Lawrence Berkeley National Laboratory, "Incorporating Data Center Specific Sustainability Best Practices into EO 13514 Implementation" — for each degree Fahrenheit of chilled-water supply temperature increase, chiller efficiency improves approximately 1–1.5 %; a constant-speed chiller reset from 44 °F to 60 °F cuts power from 0.63 to 0.40 kW/ton; a 10 °F intake-air increase in a mild climate adds about 1,300 free-cooling hours. datacenters.lbl.gov (PDF)
19. Chen Kuo Ying HVAC Engineers (Taiwan), "Data center PUE regulation and cooling design strategy" — each 1 °C of chilled-water supply reset improves chiller COP by roughly 2–3 %; the 7 → 12 → 18 → 24 °C ladder with COP gains of 15–20 % and 20–30 % per step and water-side free-cooling hours rising from under 500 to over 1,500 hours at an 18 °C supply in a subtropical climate; variable-flow pump affinity laws (80 % flow = 51 % power). cky.com.tw
20. EnerShares, "Hidden Inefficiency in Chiller Plants: Low Delta T Syndrome" — a 1,000 TR plant designed at a 5 °C ΔT circulates about 605 m³/h and 1,590 m³/h when the ΔT collapses to 1.9 °C; flow must rise inversely with ΔT for the same load, with pump energy following the cube law; typical total plant energy penalty 15–25 %. enershares.com
21. DatacentersX, "Data Center Cooling Towers and Heat Rejection" — rejection-technology approach table: open cooling tower 3–5 °C above ambient wet-bulb, closed-circuit fluid cooler 4–7 °C above wet-bulb, dry cooler 8–12 °C above dry-bulb, hybrid/adiabatic variable and toward wet-bulb in adiabatic mode; direct-to-chip supply acceptance of 30–45 °C keeps chillers off for most of the year. datacentersx.com
22. HVACR Nexus, "Heat Rejection Calculator" — rejection equipment is sized on approach, not load: going from a 10 K to a 5 K approach roughly doubles the coil surface for the same duty, below about 3 K dry cooling stops being practical at summer ambients, and raising the fluid temperature is usually the cheapest lever available; glycol lowers specific heat and raises pumping power. hvacrnexus.com
23. Blockchain Miner, "Dry Cooler Approach Temperature Explained" — a project cannot use the server coolant target as the dry-cooler leaving-fluid target without checking the complete temperature budget; margin is lost at three points (dry-cooler approach to ambient, CDU exchanger approach, secondary temperature rise through the racks); reducing approach from 10 K to 5 K is not a minor adjustment. blockchain-miner.com
24. Yobitel, "Direct-to-Chip Liquid Cooling (DLC)" — for warm-water W3/W4 operation design a 3–5 °C approach at the CDU heat exchanger, giving a facility supply of 30–37 °C sized to the local 99 %-design wet bulb; in London a 7 °C dry-cooler approach still makes 35 °C supply on the hottest day; server-side supply 25–32 °C, return 35–60 °C with an 8–15 °C cold-plate ΔT; heat capture to liquid 85–95 %. yobitel.com
25. Augmintech, "What Is a Chilled Water System?" — AHRI Standard 550/590 standard rating conditions of 44 °F (6.7 °C) leaving and 54 °F (12.2 °C) entering chilled water, outside which a manufacturer’s published efficiency no longer applies directly; the 7/12 °C convention traced to comfort-air-conditioning practice; Q = ṅ·cp·ΔT and the flow equation with a 100 TR worked example. augmintech.com