How to Size Coolant Pipe Diameter from CDU Operating Conditions

Two engineers are handed the same CDU datasheet. One orders 4 inch stainless mains, because 4 inch is what the last project used and it is a stock size. The other runs four lines of arithmetic and orders 5 inch. Both loops will run. One of them will carry 3.0 m/s through the main header — 43 % above the erosion limit that applies to pipes this large — and will pay 2.7 times the pressure drop to do it. The arithmetic is not hard. It is just usually skipped.

This article walks the full chain in the order the physics forces on you: CDU capacity, the secondary ΔT you are designing to, the flow that heat load demands, the velocity band the fluid and the pipe material allow, the diameter that velocity implies, and finally the proof that the CDU's own pump head covers the resulting pressure drop. Every number in the worked example below is calculated, not quoted.

Pipe diameter is a CDU output, not an input

In a direct-to-chip loop there is exactly one component that both creates the flow and owns the pressure available to push it: the coolant distribution unit. Everything downstream of it — the row manifold, the rack manifold, the quick disconnects, the cold plates — arrives as a fixed, vendor-specified pressure-drop package. The CDU's heat exchanger sits between two loops that must never mix, the facility water system (FWS) on the primary side and the technology cooling system (TCS) on the secondary side, and everything about the TCS side downstream of the CDU is decided before you get to it [1][2].

What you still own is the pipe. And the pipe is not a passive piece of plumbing. Its diameter sets the operating point on the CDU's pump curve. A larger main drops the loop resistance, which lets the pump deliver more flow at a lower head, which shrinks the secondary ΔT and improves cold-plate performance — while raising pump power and, at the extreme, dropping velocity so low that the loop stops carrying entrained air and particulate to its filter. A smaller main does the reverse. There is a band, not a point, and the band is set by velocity.

So the chain only runs one way:

Step 1 — Read four numbers off the CDU datasheet

Every liquid-to-liquid CDU datasheet carries the same handful of fields, and each one sizes something different. Read them as sizing inputs, not as marketing.

The most expensive misreading on that list is the difference between two flow numbers that look interchangeable.

The rated capacity flow is a test condition. When the Open Compute Project standardized how liquid-to-liquid CDU capacity should be reported, it fixed the rating point at X kW with both the TCS and FWS sides flowing at X × 1.5 L/min, an approach temperature difference of 5 K, a minimum TCS head (25 psi for in-rack units, 40 psi for in-row units), and a maximum facility-side impedance of 75 psi, using facility water at 40 °C and PG25 on the TCS side [3]. Those are laboratory conditions chosen so that two vendors' numbers can be compared. They are not your operating point.

The same body’s CDU specification guidance goes further, and it is worth reading before you size anything: it covers right-sizing a CDU for peak and partial load rather than peak alone, flow-control and pump-redundancy strategy, and concurrent maintainability — the requirement that components be serviceable without interrupting IT cooling. Every one of those requirements lands on the pipe, because a CDU tuned for partial-load efficiency and a CDU oversized for peak load demand different flow ranges through the same main [17].

The design flow is what your load and your chosen ΔT actually require. Size the pipe on the higher of the two, because the high one is what the pump can actually produce on a bad day — and check the low one against the velocity floor.

Step 2 — Turn kilowatts into litres per minute

The heat load and the flow are linked by one equation, and it is the only one in this article you cannot skip:

Two things about that equation cause most of the sizing errors in the field. The first is Q: you must use the share of rack power that actually reaches the cold plates, not the rack's nameplate. Memory, drives, power supplies, NICs and the fan wall still reject to air on most hybrid platforms, and sizing the loop on total rack power oversizes the flow — which then pushes you to a larger pipe than the coolant ever needed [4].

The second is ΔT. Flow scales inversely with it, so the difference between a 10 K and a 15 K design is a 50 % swing in flow, and therefore a full pipe size.

Run the same equation backwards and OCP's 1.5 L/min-per-kW rating basis stops looking like a round number. For PG25 it implies a secondary ΔT of 9.87 K; for water, 9.65 K. The industry's standard CDU rating point is, to within 1 %, a 10 K design. If your design ΔT is also 10 K, your real flow and the CDU's rated flow coincide and life is simple. If your design ΔT is 12 K or 15 K, they do not, and a pipe sized naively off the datasheet flow will run 17 % or 34 % slower than you calculated.

Step 3 — Choose a velocity band, not a pipe size

Velocity is the only variable in the chain that is genuinely a design decision. It is bounded on both sides, and the bounds come from different failure modes.

At the top end, sustained high velocity erodes pipe and fittings — and the mechanism matters, because it is erosion-corrosion rather than simple mechanical wear. Erosion corrosion needs chemical attack and mechanical wear at the same time, and the mechanical half comes mainly from solid particles in the fluid: more particles, lower tolerable flow rate. And the tolerable velocity is strongly material-dependent — high-alloy stainless steels tolerate more than 20 m/s, while copper and copper alloys begin to be attacked from around 1 m/s [6]. That is why ASHRAE's liquid-cooling guidance caps coolant velocity at 2.1 m/s (7 ft/s) in pipes larger than 3 inch to limit erosion of pipe material, and holds flexible tubing at or below 1.5 m/s (5 ft/s), where vibration and water hammer add to the wear [10][19]. A well-filtered loop tolerates more than a dirty one; the cap is a design rule written for the worst credible water quality, not a property of water.

Below the working band, the problem is different again: velocity low enough that the loop stops transporting entrained air to its air vents and stops carrying particulate debris into the CDU's 50 µm filter. The floor is generally taken as 0.6 m/s (2 ft/s) [7].

Notice the tension between rows two and three. Industry calculators commonly quote a 1.5–3.0 m/s band for TCS headers, but ASHRAE's liquid-cooling guidance caps pipes larger than 3 inch at 2.1 m/s on erosion grounds [6][10]. For mains at or above DN80, treat 2.1 m/s as the hard ceiling and 1.5–2.0 m/s as the working band. The 3.0 m/s figure belongs to small-bore and pump-discharge service, where the pipe wall is thick relative to the diameter and the run is short.

Step 4 — Convert flow and velocity into a diameter

With flow in cubic metres per second and velocity in metres per second, the diameter falls straight out of the continuity equation:

Take a worked case: a 1 000 kW liquid-to-liquid CDU on PG25, designed to a 10 K secondary rise, feeding a row at a TCS supply of 32 °C with facility water at 27 °C (a 5 K approach, inside the OCP rating condition).

Note what the fourth row is telling you. To land on a 4 inch steel pipe you must accept 3.0 m/s — the absolute top of the general working band and 43 % above the ASHRAE erosion ceiling that applies to pipes this size. The arithmetic is already warning you. Step 5 makes it explicit.

Step 5 — Test every candidate nominal size properly

This is the table that decides the project, and it is the one most often replaced by “we always use 4 inch.” For each candidate size, compute the actual inside diameter, the actual velocity, the Reynolds number, and the friction rate. Reynolds number matters because it tells you which regime you are in and therefore how much you can trust the friction factor:

Two independent criteria — the velocity band and the friction rate — point at the same answer here, which is what you want. The 4 inch pipe is not merely marginal; it is 43 % past the erosion ceiling and 85 % above the top of the economic friction band. The 8 inch pipe is the opposite failure: at nameplate flow it is already at 0.76 m/s, and at half load it drops through the 0.6 m/s floor and starts collecting air and debris. The 5 inch pipe sits at 1.91 m/s with 10 % headroom before it even touches the 2.1 m/s cap.

Step 6 — Cross-check with the friction rate

Velocity sizing answers “will this pipe survive the fluid.” Friction-rate sizing answers “will this pipe make the pump work too hard.” In a data center TCS loop the runs are short and the CDU's head budget is generous, so velocity usually governs — but on long facility-water runs to a remote heat-rejection plant the friction rate takes over, and it is worth running both. The classical hydronic rule, carried in the ASHRAE Handbook and in ANSI/ASHRAE Standard 90.1's piping tables, is simple: pipes 2 inch and smaller are sized to a velocity limit of 4 ft/s (1.2 m/s); pipes above 2 inch are sized to a friction loss of 4 ft per 100 ft (about 400 Pa/m); the enclosing hard limits are 2 ft/s and 15 ft/s, and 0.75 to 10 ft per 100 ft [8][9][12].

The friction factor comes from the Darcy-Weisbach framework, with the friction factor evaluated on the Haaland explicit approximation to Colebrook-White — accurate to well within the ±10 % that pipe-roughness and fitting-count uncertainty justify, and far easier to put in a spreadsheet than an implicit equation [13][14]:

For the selected 5 inch pipe the numbers come out as follows. Reynolds number 166 400, relative roughness ε/D = 1.17 × 10⁻⁴, Haaland friction factor f = 0.0168. Over a 60 m one-way main that is 14.6 kPa, or 244 Pa/m — 2.44 ft per 100 ft, comfortably inside the 1–4 ft-per-100 ft economic band. The 4 inch alternative lands at 7.38 ft per 100 ft, over the 4 ft guideline and only saved from 10 ft by the short run length. Two methods, one answer.

Step 7 — Prove the CDU's head budget covers the loop

A pipe that satisfies velocity and friction can still fail, because the real question is whether the CDU's pump can push the design flow through everything — and minor losses in a liquid-cooled loop are not minor. In a typical direct-to-chip installation the fittings, quick disconnects, strainers and heat exchangers contribute more pressure drop than the straight pipe does [10]. The standard way to account for them is the resistance-coefficient method, set out in Crane Technical Paper 410, where each fitting carries a K value measured in velocity heads. The couplings counted in that accounting are not unknowns either — the OCP Universal Quick Disconnect specifications define the pressure-drop and leakage envelope of the dry-break connectors that sit at every rack boundary, and those published envelopes are the right figures to use rather than a generic coupling estimate [18]:

Now close the loop. At 1.91 m/s the velocity head is 1 857 Pa, so ΣK = 16 costs 29.8 kPa. Straight-run friction over 60 m costs 14.6 kPa. Total: 0.444 bar, or 6.4 psi.

Compare that against the head the CDU is required to produce. OCP's capacity-rating conditions set a minimum TCS head of 25 psi at the rated flow for an in-rack CDU and 40 psi for an in-row CDU [3]. So the main piping consumes 26 % of an in-rack unit's rated head and 16 % of an in-row unit's, leaving 18.6 or 33.6 psi for the row manifold, the rack manifolds, the quick disconnects and the cold plates themselves — a range that comfortably brackets the typical 0.5–3.0 bar (7–44 psi) cold-plate differential-pressure window published by CDU manufacturers [10].

Then run the same arithmetic on the 4 inch alternative, and the case closes hard. At 3.00 m/s the velocity head rises to 4 582 Pa, so the same fittings cost 73.3 kPa, and friction over the same 60 m costs 44.3 kPa. Total 1.18 bar, or 17.1 psi — 2.65 times the 5 inch figure, and 68 % of an in-rack CDU's entire rated head, spent before a single cold plate is connected. That is the real cost of the “we always use 4 inch” decision: not the pipe, the pump margin.

Step 8 — Size the primary side separately

The FWS side is a different design problem and must not inherit the TCS pipe size. Three things change. The fluid is usually plain water or a diluted glycol, so the flow for the same heat load is slightly lower. The approach temperature defines the supply temperature you must deliver rather than the temperature you must hold. And the impedance budget is stated separately: OCP's rating conditions allow a facility-side impedance of up to 75 psi at the rated flow, which is a much larger number than the TCS head because the primary side has to reach the building plant [3].

Two useful things fall out of that table. First, because water's density and specific heat are both higher than PG25's, the primary flow for the same heat load is about 2 % lower than the secondary flow — close enough that engineers routinely match the two sides, but worth calculating rather than assuming. Second, watch the friction factor column: it barely moves across a 2.4 times range of velocity. At these Reynolds numbers you are deep in the roughness-dominated regime, which means the primary lever on pressure drop is diameter, not flow rate, and not the difference between water and glycol.

Where redundancy changes the answer

Redundancy is usually treated as an electrical problem. In a pumped loop it is a hydraulic one, and it changes the pipe diameter, because parallel mains divide the flow.

The middle row is the trap. If both 2N paths are running and sharing load, a 4 inch main looks perfect at 1.50 m/s — and then one path is isolated for maintenance and the survivor carries double the flow at 3.00 m/s, 43 % over the erosion cap, for the duration of the outage. Redundant piping must be sized for the failover case, not the normal case. Size on full flow through one path, and treat the shared-load velocity as a part-load floor check, not as your design point.

Two lookup tables to skip the arithmetic

For early-stage work, the two tables below cover most questions. The first expresses the ASHRAE erosion ceiling as a capacity limit: how much CDU load a given main can serve before the pipe is the constraint.

The second is the reverse lookup: the flow each nominal size can carry while staying inside the working velocity band.

Eight mistakes that get the diameter wrong

1. Sizing to nominal size instead of inside diameter. A 4 inch steel pipe is 102.3 mm inside; a 4 inch Type L copper tube is 99.2 mm; a 4 inch Schedule 80 pipe is 97.2 mm. On a 4 inch line that is a 5 % diameter spread — and because pressure drop scales roughly with the inverse fifth power of diameter, a 5 % error in diameter is a 28 % error in pressure drop [12].
2. Using rack nameplate kW instead of the liquid-cooled fraction. Memory, drives and power supplies still reject to air on most hybrid platforms. Sizing on total rack power oversizes the loop and, worse, pushes the velocity below the transport floor.
3. Designing at the wrong ΔT. A 15 K design needs 33 % less flow than a 10 K design. If the pipe is sized at 10 K and the loop is commissioned at 15 K, every velocity in the calculation is wrong by a third.
4. Assuming glycol costs 17 % more flow. At 25 wt % the volumetric penalty is 2.3 %. Applying the 50 % figure buys pipe you do not need and buys it at the cost of a velocity that is now too low.
5. Ignoring minor losses. In a direct-to-chip loop the fittings, quick disconnects and strainers often exceed the straight-pipe friction. In the worked example, ΣK = 16 accounted for two thirds of the total pressure drop while 60 m of pipe accounted for one third.
6. Using a single K value for a control valve. Fitting K values apply to fully open valves. A partially open or modulating valve must be sized on its flow coefficient, because its resistance rises steeply with travel [15].
7. Sizing N+1 and 2N piping on the shared-load case. Redundant mains must be sized on full flow through one path, or the survivable failure becomes an erosion event.
8. Forgetting the fouling and aging allowance. A 1.10 to 1.20 safety factor on head loss is normal engineering practice, because real systems have fouling, aging pumps and an uncertain fitting count. Straight-run friction in steel can rise 20–50 % over the life of a poorly treated system as roughness increases [12].

What the pipe cannot fix: dew point

There is one constraint that has nothing to do with diameter and everything to do with whether the loop ruins the room. The TCS supply temperature must stay above the room dew point with margin — ASHRAE's liquid-cooling guidance points to at least 2 to 3 K — or condensate forms on cold plates, on manifolds, and on every uninsulated surface in the loop [1][10]. That is a temperature and insulation problem, and no amount of oversizing the pipe addresses it.

It is, however, the constraint that sets the FWS supply temperature you are allowed to use, and therefore the approach you can afford, and therefore the CDU you should be specifying in the first place. The whole chain — dew point margin, FWS class, CDU approach, TCS supply, secondary ΔT, flow, diameter — hangs together from the room's air conditions outward. We worked the dew-point side of that chain in full, including the ASHRAE saturation-pressure polynomial and a 7 × 6 calculated dew-point grid, in How to calculate dew point in a data center.

Material, pressure class and cleanliness

Diameter is only useful if the pipe survives the fluid and the pressure. Four points matter for the loop you just sized.

Pressure class. OCP's liquid-to-liquid CDU guidance notes that the maximum pressure supplied to a technology cooling system could be as high as 100 psi (690 kPa), and that every ingredient in the TCS should be designed to meet it [1]. That is well above what a Schedule 40 line needs structurally, but it drives flange ratings, joint types, and quick-disconnect selection. ASME B31.9, Building Services Piping, is the code that normally governs this class of in-building loop in the United States [16].

Wetted materials. ASHRAE's liquid-cooling guidance is specific about what may touch the fluid: copper; brass with less than 15 % zinc and no lead; 304L or 316L stainless steel; peroxide-cured EPDM elastomer. And it is explicit about what must never be used in a closed loop: aluminium, brass with more than 15 % zinc, lead-containing brass, and non-stainless ferrous metals [1][6][19].

Filtration. Debris is normally caught at the CDU with filter elements rated at 50 µm or finer, but a filter only works if debris reaches it. That is the physical reason for the 0.6 m/s velocity floor, and it is why an oversized main is a fouling risk rather than a safety margin [10].

Cleanliness. Cold-plate microchannels are the narrowest and most fragile part of the entire system. Cleanliness targets are written against ISO 4406 or NAS 1638 codes, and loops are flushed until the particulate count and conductivity stabilise before any cold plate is connected [10].

Commissioning checks that close the loop

1. Measure the actual flow at the CDU's TCS outlet and compare it against the design flow calculated in Step 2 — not against the datasheet rating.
2. Measure the secondary ΔT at design load and back-calculate the real liquid-cooled heat load. If the ΔT is lower than design, the flow is higher than design and every velocity in your table is optimistic.
3. Measure the differential pressure across the main run at design flow and compare against the calculated value with the 1.10–1.20 margin. A large positive surprise means a fitting, a strainer or a partially closed valve is costing you head.
4. Verify the CDU's pump curve against the as-built system curve rather than the design curve — the pipe you installed, not the one you drew.
5. Prove the failover: isolate one path of the redundant pair and confirm the survivor stays inside the erosion cap at full load.
6. Confirm the TCS supply temperature holds at least 2–3 K above the measured room dew point at the worst humidity condition of the year.

The short version

Pipe diameter in a CDU-fed liquid loop is not a convention and it is not a stock-size habit. It is the output of four numbers: the liquid-cooled heat load, the secondary ΔT you are designing to, the velocity band your fluid and pipe material allow, and the head your CDU can actually produce. Get those four and the diameter falls out of two equations.

In the worked example, the habit answer and the calculated answer differed by one nominal size and by 2.65 times the pressure drop. On a single CDU that is a pump-margin problem. Across a hall of them, with a fixed facility-water supply temperature and a fixed cold-plate differential-pressure window at the far end, it is a decision about whether the loop delivers design flow at all.

If you are laying out the fluid path for a high-density row — rack manifolds, quick disconnects, manifold sizing and the enclosure geometry around them — the arithmetic above is the specification your hardware has to fit. Send us the CDU operating conditions and the rack layout and we will work the flow, velocity and manifold sizing through with you.

References

1. ASHRAE Technical Committee 9.9, Water-Cooled Servers: Common Designs, Components, and Processes, white paper, 2019 — CDU and non-CDU architectures, the FWS/TCS split, the 100 psi (690 kPa) maximum supply pressure for the technology cooling system, wetted materials and materials to avoid, filtration, and the flexible-hose velocity guidance. ashrae.org (PDF)
2. ASHRAE Technical Committee 9.9 — committee scope, publications and the Datacom Encyclopedia, which has replaced the ASHRAE Datacom Series (including Liquid Cooling Guidelines for Datacom Equipment Centers, 2nd edition) and the Thermal Guidelines for Data Processing Environments. tpc.ashrae.org
3. Open Compute Project, “Liquid to Liquid CDU Test Methodology and Performance Rating,” Revision 1.0, August 2024 — the approach temperature difference (ATD) definition, and the proposed standard capacity-rating conditions: facility water at 40 °C (ASHRAE W40), PG25 on the TCS side, TCS head ≥ 25 psi (in-rack) or ≥ 40 psi (in-row) at X × 1.5 L/min, facility impedance ≤ 75 psi, and ATD = 5 °C at X kW. opencompute.org
4. Data center liquid cooling engineering guide — the steady-flow energy balance Q = ṁ c_p ΔT applied to cold-plate heat removal, with a worked example (1 200 W removed at a 5 K coolant rise requires about 3.44 L/min of water), and the field warning that sizing a loop from total rack power while ignoring the fraction still rejected to air by memory, power supplies and other components is a common design mistake. mechtics.com
5. Propylene glycol 25 % solution thermophysical properties, cross-checked against two independent published tables. An independent laboratory table for a 25 % aqueous propylene glycol solution gives, at 40 °C, ρ = 1011 kg/m³, c_p = 4.00 kJ/kg·K, dynamic viscosity 1.42 × 10⁻³ Pa·s and thermal conductivity 0.491 W/m·K, across the range −10 to 100 °C. A direct-to-chip coolant product data sheet (“Castrol ON Direct Liquid Cooling PG 25”) gives, at 40 °C, ρ = 1018 kg/m³, kinematic viscosity 1.5 cSt, c_p = 3.98 kJ/kg·K and k = 0.483 W/m·K. The two differ by 0.2 % in the resulting design flow. masterchem.ee
6. Alleima (formerly Sandvik Materials Technology), “Facts about erosion corrosion” — the mechanism behind velocity limits in wetted pipe: erosion corrosion requires simultaneous chemical attack and mechanical wear, the mechanical contribution comes mainly from solid particles in the fluid, and increasing particle content reduces the tolerable flow rate. The tolerable velocity is strongly material-dependent, with high-alloy stainless steels tolerating more than 20 m/s while copper and copper alloys are attacked from around 1 m/s. alleima.com
7. ASHRAE guidance on hydronic pipe sizing as summarised in the engineering literature — preliminary sizing for a friction loss between 80 and 450 Pa per metre of straight pipe with general-service velocity not exceeding 3 m/s, the air-management recommendation to keep velocity above 0.6 m/s so entrained air is carried through the system, the copper piping trade association maxima of 1.5 m/s for hot-water and 2.4 m/s for cold-water piping, and the recommendation of no more than 1.2 m/s in pipes under 38 mm. ingener.by
8. ASHRAE. ASHRAE Handbook — Fundamentals, and ASHRAE Handbook — HVAC Systems and Equipment, hydronic heating and cooling system design — the friction-loss charts and velocity tables from which the chilled-water sizing rule of thumb is drawn: 4 ft/s (1.2 m/s) for pipes 2 inch and below, 4 ft per 100 ft (about 400 Pa/m) for pipes above 2 inch, with enclosing limits of 2 to 15 ft/s and 0.75 to 10 ft per 100 ft. ashrae.org
9. ANSI/ASHRAE Standard 90.1, Section 6.5.4, Table 6.5.4.5 — prescriptive maximum flow rates for chilled-water and condenser-water piping by nominal pipe size and annual operating hours, distinguishing variable-flow from constant-flow service; for example, 4 inch pipe is limited to 530 gpm in variable-flow service under 2 000 hours per year and 320 gpm at 4 400 hours and above. deppmann.com (PDF)
10. “How to design piping systems for data centers that require liquid cooling,” Consulting-Specifying Engineer — reports the ASHRAE liquid-cooling guidance limiting coolant velocity to 2.1 m/s for pipe diameters larger than 3 inch to minimise material erosion; the water-quality requirements for the TCS loop; CDU filtration at 50 µm or finer; and the practice of isolating each rack with valves at the in-rack manifold. csemag.com
11. ASTM B88, Standard Specification for Seamless Copper Water Tube — Types K, L and M dimensions. A 4 inch Type L tube has a 4.125 inch outside diameter and a 3.905 inch (99.2 mm) inside diameter, against 102.3 mm for 4 inch Schedule 40 steel. astm.org
12. Chilled-water pipe sizing reference tables and worked examples applying the ASHRAE rule of thumb, Schedule 40 steel properties and Type L copper properties — including the note that Schedule 40 4 inch pipe has a 102.3 mm inside bore, that pressure drop varies approximately with the inverse fifth power of diameter so a 5 % diameter error causes about a 28 % pressure-drop error, and that ageing and fouling can raise pressure drop 20–50 % over the life of an open or poorly treated system. mepbase.com
13. Haaland, S. E., “Simple and Explicit Formulas for the Friction Factor in Turbulent Pipe Flow,” Journal of Fluids Engineering, Vol. 105, No. 1, March 1983, pp. 89–90, doi:10.1115/1.3240948 — the explicit approximation to the Colebrook-White friction factor used in this article, valid for 3 × 10³ ≤ Re ≤ 1.5 × 10⁸ and 0 ≤ ε/D ≤ 5 × 10⁻². osti.gov
14. Darcy-Weisbach framework, friction-factor procedures and representative roughness values used in the worked example — commercial or welded steel ε = 0.046 mm, drawn stainless steel ε = 0.015 mm, PVC ε = 0.0015 mm; plus the note that a 5 % reduction in inside diameter (Schedule 40 versus Schedule 80 4 inch pipe) raises pressure drop by about 28 % because ΔP ∝ 1/D⁵. multicalci.com
15. Crane Technical Paper No. 410, Flow of Fluids Through Valves, Fittings, and Pipe — the resistance-coefficient method used to build Table 6, expressed as K = n · f_T where n is a fixed L/D equivalent for the fitting type (30 for a 90° standard elbow, 16 for a long-radius or 45° elbow, 20 for a tee through-run, 60 for a tee branch, 8 for a gate valve, 340 for a globe valve) and f_T is the fully-turbulent friction factor for that pipe size, so the same fitting has a larger K in small pipe than in large. simupipe.com
16. ASME B31.9, Building Services Piping — the code section containing design, materials, fabrication, installation, inspection, examination and testing rules for piping in industrial, institutional, commercial and public buildings, which is the class most in-building data center hydronic loops fall under in the United States. asme.org
17. Open Compute Project, “Qualification & Performance Specification Guidance for CDUs in Modular Technology Cooling Systems,” draft — CDU chemistry and fluid-quality control including integrated filtration and air/gas removal, flow-control and redundancy requirements, pressure and flow performance specification, right-sizing for peak and partial load, and concurrent maintainability. ocp-all.groups.io (PDF)
18. Open Compute Project, Cold Plate and Coolant Distribution Unit sub-projects — accepted contributions including Guidelines for Using Water-Based Transfer Fluids in Single-Phase Cold Plate-Based Liquid-Cooled Racks, Guidelines for Using Propylene Glycol-Based Heat Transfer Fluids in Single-Phase Cold Plate-Based Liquid Cooled Racks, Cold Plate Cooling Loop Requirements Rev. 2, and the Universal Quick Disconnect specifications that define the pressure-drop and leakage envelope of the couplings counted in Table 6. opencompute.org
19. Peerless Electronics, “5 things to consider when selecting connections for liquid cooling servers,” citing Liquid Cooling Guidelines for Datacom Equipment Centers, 2nd edition — the wetted-material recommendations (copper; brass under 15 % zinc and lead-free; 304L or 316L stainless; peroxide-cured EPDM), the materials to avoid, and the requirement that flexible tubing be maintained below 5 ft/s (1.5 m/s) to control erosion, vibration and water hammer. peerless-inc.com