Pressure Drop Piping Turns 4x Flow Into 52x Hydraulic Power

The OCP OAI liquid cooling guidelines report a cold plate server loop whose pressure drop, measured from quick disconnect to quick disconnect, reached 17 psi at 8 L/min. Roughly half of that went to hoses, a manifold and manually mated quick disconnects, not to the cold plates. That joint between server and rack is where added flow gets expensive, because the pump upstream has to supply all the pressure it consumes. This piece follows how pressure drop and pump power climb as flow rises, using published pipe, manifold and server-loop figures, and leaves out cold plate thermal design, two phase loops and pump selection because the cited documents print no pump curves.

Quadrupling flow in an 8-inch glycol loop raised friction loss 13 times and hydraulic power 52 times

A Specifying Engineer case study of a Midwest data hall prints pressure drop piping figures for one run of 8-inch schedule 40 steel pipe carrying 40% ethylene glycol, at normal flow on day one, normal flow at the ultimate design, and the ultimate flow in one segment while its twin is isolated.

380 gpm: 0.3 ft water column per 100 ft at 2.4 ft/s.

760 gpm: 1 ft water column per 100 ft at 4.9 ft/s.

1,520 gpm: 3.9 ft water column per 100 ft at 9.8 ft/s.

Over the full 4× span the drop rises 13 times, which fits a power law with exponent n = ln 13 ÷ ln 4 = 1.85.

The first doubling alone gives n = 1.74, soft because the starting 0.3 ft is printed to one digit. The second gives 1.96.

Pressure follows from head, ΔP = ρ × g × h. A foot of water column is a water-equivalent pressure, so the density of water applies even though the loop carries glycol. The calculation takes water at 30 °C as the reference, for which NIST lists 995.69 kg/m³.

380 gpm: Q = 0.02397 m³/s (1 gpm = 6.309 × 10⁻⁵ m³/s); h = 0.0914 m (1 ft = 0.3048 m); ΔP = 893 Pa; P = Q × ΔP = 21 W per 100 ft.

1,520 gpm: Q = 0.09590 m³/s; h = 1.189 m; ΔP = 11,607 Pa; P = 1,113 W per 100 ft.

Flow rose 4 times, the drop 13 times and the hydraulic power 52 times, which is 4 × 13.

Pump shaft power exceeds hydraulic power by the pump and motor efficiency, which the case study does not state, so 1,113 W is a lower bound on electrical input for that pipe.

A pipe checked only against a velocity limit passes the isolation case while drawing 52 times the base-case friction power for as long as the segment stays out of service.

A flow network model and a bench prototype differ by 2.4 to 2.7 times on 4 × 2 cold plate loops

The OCP OAI System Liquid Cooling Guidelines give two pressure-drop figures for a network of 4 parallel × 2 series cold plates, from different contributors, and they do not agree.

A flow network model with 25% propylene glycol (PG25) coolant gives 10 psi at 10 L/min. The model was calibrated against test data from an earlier 4 × 2 prototype.

A separate prototype run with deionized water measured 17 psi at 8 L/min, quick disconnect to quick disconnect.

Exponent from the model’s 8 parallel × 1 series layout: 7 psi at 10 L/min and 10 psi at 12.5 L/min give n = ln (10 ÷ 7) ÷ ln (12.5 ÷ 10) = 1.60.

Scaling the 4 × 2 model to 8 L/min: 10 psi × (8 ÷ 10)^n = 7.0 psi for n = 1.6 and 6.4 psi for n = 2. The bench reading is 17 ÷ 7.0 = 2.4 and 17 ÷ 6.4 = 2.7 times higher.

The coolant does not explain the gap, since the guidelines note that viscosity falls as the glycol share falls and deionized water should therefore read lower.

The hardware is the likelier cause. Half of the measured drop is 8.5 psi, within 1.5 to 2.1 psi of the scaled model, which suggests the model covers little beyond the cold plates.

The guidelines do not state the model’s boundary, so that reading is an inference. The OCP Open Rack V3 Blind Mate Manifold Specification draws the same boundary from the other side: its curve for a manifold pair leaves out blind mate valves, passive cold plate loops, rack-level hoses and large manual quick connects.

Source: Open Compute Project, OAI System Liquid Cooling Guidelines, sections 3.4, 6.1 and 6.3.3

A pressure budget built from cold plate curves alone starts well short at this flow.

Choosing a 7.5 °C coolant rise over 12 °C costs 1.6 times the flow and 3.4 to 4.1 times the hydraulic power

Flow set at the server loop is what the pump in the coolant distribution unit that drives the secondary loop has to deliver.

The OAI guidelines recommend a coolant temperature rise between 7.5 °C and 12 °C for PG25 coolants.

That range equals 1.25 to 2.0 L/min per kW.

The typical design target is 10 °C, or 1.5 L/min per kW.

Heat balance check: Q = P ÷ (ρ × c_p × ΔT), with ρ = 995.69 kg/m³ and c_p = 4.1796 kJ/(kg·K) for water at 30 °C.

For 1 kW, Q = 1.20, 1.44 and 1.92 L/min at ΔT = 12, 10 and 7.5 °C.

Each OCP figure therefore sits 4% above the water value.

Moving from the 12 °C end of the range to the 7.5 °C end multiplies flow by 2.0 ÷ 1.25 = 1.6.

The pressure exponent is taken between 1.6 for the cold plate network and 2 as the upper bound for a pipe, which brackets the 1.85 fitted to the case study.

Pressure rises 1.6^1.6 = 2.1 to 1.6^2 = 2.6 times.

Hydraulic power, Q × ΔP, rises 1.6^2.6 = 3.4 to 1.6^3 = 4.1 times.

Hydraulic power of the 17 psi loop: 8 ÷ 60,000 m³/s × 117,211 Pa = 15.6 W.

At 1.5 L/min per kW that flow carries 5.3 kW, so 15.6 W is 0.29% of the heat moved.

The guidelines frame a higher rise as a PUE gain paid for in cooling performance, and the arithmetic leaves pump power as the smaller cost of a low rise.

Velocity ceilings, drop budgets and flow ratios come from different sheets and cannot be stacked into one pipe size

Four velocity figures bear on a coolant pipe, and each was written for a different component.

Table 1. Flow a 150 mm bore carries at four stated velocity limits (L/min; kW)

* Flow = v × π/4 × (0.150 m)², bore assumed equal to a nominal 150 mm; real bores differ. Heat uses 1.5 L/min per kW, the OCP OAI typical target for PG25 coolant at a 10 °C rise, applied here to a pipe it was not written for.

Source: LBNL working group draft (2018-06-20); OCP rack manifold white paper; Specifying Engineer (2019-09-17 and 2025-04-08); OCP OAI System Liquid Cooling Guidelines. Flow and heat columns calculated for this article.

The 2.1 m/s figure reaches this piece through Specifying Engineer, which attributes it to an ASHRAE TC 9.6 white paper that was not opened here.

The case study’s 9.8 ft/s is called within recommended limits, while the ASHRAE-attributed ceiling is 6.9 ft/s. One is a facility chilled water loop and the other a technology cooling loop, and neither article reconciles them. A sheet that quotes a velocity limit without naming the loop it covers cannot size a pipe in either.

The LBNL-led working group draft sets a different kind of limit. It caps the pressure drop of a manifold’s supply and return at less than 1/40th of an individual server loop’s drop at maximum flow.

Source: Open Standards Harmonization Working Group, Open Specification for a Liquid Cooled Server Rack, Progress Update, “Manifold Design Considerations”

On the 17 psi loop above, 1/40th is 0.43 psi.

The ORv3 manifold specification calls for a minimum branch pressure drop, set by computational analysis and on the order of 1 to 2 psi for a manifold with a reasonably large flow section, to promote uniform flow distribution. The imbalance this guards against is the subject of this site’s analysis of flow distribution across a rack manifold.

Pressure drop in these sources climbs near the square of flow, so hydraulic power climbs near the cube. The sources disagree on how much of a server loop’s drop sits outside the cold plates, with model and bench readings 2.4 to 2.7 times apart, and on how fast coolant may travel, with ceilings that differ by component. Coolant loop design that raises flow therefore needs a pressure budget built from connector-inclusive measurements and a velocity check against the specific component, and it needs qualified engineering review before any number here is used.

References

• Open Compute Project, OAI System Liquid Cooling Guidelines / OCP / file name dated 2023-03-03 (update) / https://www.opencompute.org/documents/oai-system-liquid-cooling-guidelines-in-ocp-template-mar-3-2023-update-pdf
• Open Compute Project, Open Rack V3 Blind Mate Manifold Specification, Revision 1.0 / OCP / 2024-04-05 / https://www.opencompute.org/documents/open-rack-v3-blind-mate-manifold-specification-rev-1-0-review-april05-2024-pdf
• Open Compute Project, White Paper: Guidelines to Rack Manifold Requirements and Qualification / OCP (Intel and contributors) / undated PDF / https://www.opencompute.org/documents/ocp-white-paper-rack-manifold-requirements-and-qualification-v3-pdf
• Open Standards Harmonization Working Group, Open Specification for a Liquid Cooled Server Rack, Progress Update (draft) / LBNL and members / 2018-06-20 / https://datacenters.lbl.gov/sites/default/files/OpenSpecification.pdf
• Case study: Data center piping / Specifying Engineer (Consulting-Specifying Engineer) / 2019-09-17 / https://www.csemag.com/case-study-data-center-piping/
• How to design piping systems for data centers that require liquid cooling / Specifying Engineer / 2025-04-08 / https://www.csemag.com/how-to-design-piping-systems-for-data-centers-that-require-liquid-cooling/
• Isobaric Properties for Water (P = 2 bar, 20 to 40 °C) / NIST Chemistry WebBook, SRD 69 / undated / https://webbook.nist.gov/cgi/fluid.cgi?P=2&TLow=20&THigh=40&TInc=10&Digits=5&ID=C7732185&Action=Load&Type=IsoBar&TUnit=C&PUnit=bar&DUnit=kg%2Fm3&HUnit=kJ%2Fkg&WUnit=m%2Fs&VisUnit=uPa*s&STUnit=N%2Fm&RefState=DEF