Liquid Cooling Standards Fix the Coupling Fit, Not the Flow

The Open Compute Project’s Universal Quick Disconnect (UQD) specification v2.0.0 accepts a UQD04 coupling whose flow coefficient, Cv, falls anywhere between 1.1 and 1.6. Parts at either end of that window carry the same UQD04 v2 marking and go through the same cross-verification between suppliers. That window is where liquid cooling standards stop short of interoperability in the hydraulic sense: two couplings that mate cleanly need not resist flow the same way. The coolant inside the loop has even less pinned down.

UQD v2 standardizes the mating face and leaves the pressure drop curve open

The OCP document, effective September 22, 2026, merges the separate Revision 1.0 texts for the hand-mated UQD and the blind-mated UQDB into one specification. Its scope is couplings in non-combustible single-phase water or glycol systems, and its overview says it “defines interface dimensions and performance criteria for supplier interoperability.”

The fixed part is substantial. Socket and plug dimensions are tabulated for each size, the plug tip and guide cone must be one of the listed stainless steel grades, and handmate parts need red and blue identification. Every size shares a maximum operating pressure of 6.89 bar(g) and a minimum burst pressure of 20.68 bar(g).

The open part sits in the same Table 6. The row for the pressure drop coefficients A and B, used in the reference curve ΔP(Q) = A × Q² + B × Q, has no minimum or maximum; both cells say definition is planned for a later revision. Suppliers report the curve, but nothing bounds it except a Cv floor and ceiling. Maximum flow is likewise left to each supplier, with only a general recommendation tied to a 5 m/s velocity.

Cross-verification is real but narrow. Each supplier must test against at least three others, using a subset of the property tests: fluid loss, force to connect, pressure drop and seal washout. Testing across versions is excluded. The appendix warns that v1 couplings may show higher pressure drop than v2 and that some v1 sets may not fully couple because of internal tolerance stacks. A row of v1 couplings already in service therefore has no tested path to v2 parts, and the specification hands that assessment to the end user.

Two compliant UQD04 couplings can differ by a factor of 2.1 in pressure drop

The specification reports Cv with flow in GPM and pressure drop in psi, so for water the relation rearranges to ΔP = (Q / Cv)². The assumptions are water at a specific gravity of 1, one mated pair, and the UQD04 rated flow of 1.7 GPM (6.44 L/min) from Table 6.

At the floor, Cv 1.1: ΔP = (1.7 / 1.1)² = 1.545² = 2.39 psi, or 2.39 × 6.895 = 16.5 kPa.

At the ceiling, Cv 1.6: ΔP = (1.7 / 1.6)² = 1.0625² = 1.13 psi, or 1.13 × 6.895 = 7.8 kPa.

Ratio: 2.39 / 1.13 = 2.1.

These are bounds implied by the window, not measured values. A server branch with one supply pair and one return pair doubles the coupling drop: 2 × 2.39 = 4.78 psi at the floor against 2 × 1.13 = 2.26 psi at the ceiling.

Published UQD04 values from three manufacturers all fall below 1.4, inside the window, and each sheet states its test basis differently. Fitok gives a value for each flow direction; the others give one.

Table 1. UQD04 flow coefficients and the pressure drop they imply (Cv; psi at 1.7 GPM)

* Calculated as (1.7 / Cv)² for water at specific gravity 1, one mated pair; not a measured value. Manufacturer Cv values use differing test bases.

Source: Open Compute Project, “Universal Quick Disconnect (UQD) v2.0.0” (September 22, 2026); FITOK, “Liquid Cooling Universal Quick Disconnect UQD Series” (undated); Amphenol Industrial, “UQD UQDB Catalog” (undated); Danfoss, “Danfoss Hansen Universal Quick Disconnect (UQD)” fact sheet (September 2026)

Amphenol’s catalog claims conformance to Revision 1.0 and lists its working pressure as 0 to 0.6 MPa, below the 6.89 bar(g) that v2 sets. A v2 rating cannot be read into a part that cites the older text.

The imbalance this can cause has a simple upper bound. If two branches hang off the same manifold differential and the couplings were their only resistance, flow would scale with Cv, so the low branch would get 1.1 / 1.6 = 0.69 of the high branch’s flow. Cold plates and hoses add series resistance and shrink that gap, which is why branch pressure drop governs how much manifold imbalance a rack absorbs. For a rack that mixes coupling suppliers across branches, Cv belongs on the bill of materials beside the part number, because the UQD04 label does not carry it.

The term “rated flow” also drifts. Specialty Manufacturing’s UQD04 sheet cites Revision 1.0 and prints a rated flow of 2.2 GPM, against 1.7 GPM in v2.

Evaluated at 2.2 GPM, the same Cv 1.1 coupling drops (2.2 / 1.1)² = 4.00 psi, which is 4.00 / 2.39 = 1.67 times, or 67% more than, its drop at 1.7 GPM.

A pressure drop quoted at rated flow means little until the sheet says which rated flow.

A second supplier’s PG25 counts as a different fluid under OCP’s own guideline

OCP’s propylene glycol guideline, Revision 2.0 dated September 15, 2026, covers the technology cooling system between the coolant distribution unit and the cold plates. It describes itself as a baseline of recommended practices, not rigid requirements, and its table of typical PG25 properties carries a footnote stating the values are representative and not specifications.

The ranges it gives are wide enough to admit different chemistries. Propylene glycol runs 24.5 to 27.5% by volume for PG25. Fluid pH has separate bands for OAT-based and non-OAT inhibitor packages, and the guideline says pH depends on the inhibitor formulation.

Three clauses then tie the loop to whoever filled it. The refractive index to freeze point correlation may vary from one manufacturer to another, so a field refractometer reading has to be read against that supplier’s data. Inhibitor top-ups should use only additives approved by the original fluid supplier. And equipment that holds another fluid, with an alternative PG25 formulation named as an example, should be flushed and filled with the operating fluid before it joins the loop.

The fluid supplier chosen at commissioning is effectively a single-source decision for the life of the loop. Every server or cold plate assembly that arrives holding a different PG25 is a flush event, and the guideline treats it that way. The broader point about what each coolant data sheet assumes applies here with a contractual edge.

Responsibility is shared on paper. Compatibility testing can be done by the material supplier, the component maker, the fluid supplier, the operator or an outside laboratory, and final responsibility rests jointly with manufacturers, fluid suppliers, contractors and operators. With five possible testers and no named owner, the operator, as the party holding the loop after handover, carries the gap by default.

Interoperability ends at the coupling face and the operator owns the rest

The OCP cooling documents read here settle fit, a pressure rating, a Cv window, a stainless steel requirement and a list of commonly compatible materials. They leave hydraulic resistance inside the window, mixing of v1 and v2 couplings, maximum flow and fluid chemistry beyond representative ranges to whoever writes the purchase order. The cost of that boundary shows up as pressure budget the CDU must cover, flush events when new equipment arrives, and qualification tests the operator either buys or skips.

The arithmetic has limits. It uses water at a specific gravity of 1; the specification itself says Cv is reported relative to water and that other liquids need correction factors or testing, so PG25 drops would differ from the figures above. The manufacturer values come from sheets with different or unstated test conditions and were not measured for this piece. The 0.69 flow ratio is an upper bound with the couplings as the only resistance. The Revision 1.0 coupling specification itself could not be obtained, so its rated flow and Cv minimum are not quoted. Any change to a live loop needs qualified engineering review.

A UQD04 line item or a PG25 order should come with five answers in writing:

• The coupling’s Cv at v2 rated flow in each direction, with test fluid and temperature
• The specification version each coupling claims, v1 or v2
• The supplier’s published cross-verification list of at least three other suppliers
• The fluid supplier’s refractive index chart and its approved inhibitor top-up product
• The fluid already inside any server or cold plate assembly before it is connected

References

• Cooling Environments: Cold Plate Sub-Project Design Specification, Universal Quick Disconnect (UQD) v2.0.0 / Open Compute Project / September 22, 2026 / https://www.opencompute.org/wp-content/uploads/2026/10/OCP-Design-Spec_UQD_v2.0.0.pdf
• Guidelines for Using Propylene Glycol-Based Heat Transfer Fluids in Liquid Cooled Racks, Revision 2.0, Version 1.0.0 / Open Compute Project / September 15, 2026 / https://www.opencompute.org/wp-content/uploads/2026/10/OCP-PG25-Guidelines-V1.0.0.pdf
• Liquid Cooling Universal Quick Disconnect UQD Series / FITOK / undated / https://www.fitok.com/tmp/file/Liquid_Cooling_Universal_Quick_Disconnect_UQD_Series_EN.pdf
• UQD UQDB Catalog, Data Center Liquid Cooling Solutions / Amphenol Industrial / undated / https://www.amphenol.co.jp/military/catalog/UQD_UQDB.pdf
• Danfoss Hansen Universal Quick Disconnect (UQD) fact sheet, AI423953312825en-000106 / Danfoss Power Solutions / September 2026 / https://assets.danfoss.com/documents/latest/250343/AI423953312825en-000102.pdf
• Universal Quick Disconnect (UQD) UQD04 spec sheet, Rev B / Specialty Manufacturing / September 28, 2026 / https://www.specialtymfg.com/wp-content/uploads/UQD-SpecSheet-Website.pdf