How Much Coolant Does a Quick Disconnect Actually Lose Per Mate: Back-Calculating Cv, Mating Force and Pump Power from UQD Fluid Loss

What size quick disconnect does a 700 W GPU cold plate need? Several selection guides answer 13.4 L/min and recommend UQD06. Rebuilding the unit chain shows that 13.4 is grams per second, not litres per minute: real volumetric flow is only 0.85 L/min, UQD02 is sufficient, and the connector has been oversized by a factor of five. Oversizing is not merely a purchasing penalty. Mating force at line pressure rises from 78.9 N to 200.7 N, the blind-mate mechanism must be redesigned, and the pump power budget is wrong by an order of magnitude.

This article does not compare brands. It links the four numbers published in the OCP UQD (Universal Quick Disconnect) specification into one reproducible chain: fluid loss per cycle, Cv flow coefficient, rated flow, and the rate at which mating force grows with line pressure. Every worked example supplies its inputs and intermediate values so a reader can check each line with a calculator.

Scope of the assessment: This article covers hydraulic and mechanical checks at the coupling itself: flow unit chain, Cv pressure drop, effective pressurized area, mating force, dry-break poppet timing, seal compression, and leak-detection scale. It does not cover long-term coolant degradation, particulate contamination, or rack-level flow distribution. Coolant compatibility and crevice chloride concentration are quantified in our 2026-10-26 page, branch flow uniformity in the 2026-09-24 and 2026-10-14 pages. These are separate checks and cannot substitute for one another.

1. Four hard limits from the specification

OCP UQD specification v2.0.0 (2026, consolidating the earlier hand-mate UQD and blind-mate UQDB documents into one) covers sizes 02, 04, 06 and 08 for single-phase water and glycol coolants. The published limits relevant here are listed below.

Other stated boundaries: maximum operating pressure 6.89 bar(g), minimum burst pressure 20.68 bar(g), operating temperature 5 to 65 degrees Celsius, at least 5000 mating cycles, and the plug tip, shaft and guide area must be SUS 303, 304, 304L, 316 or 316L stainless steel. Verification includes a hydrostatic hold at 6.9 bar for at least three minutes with no visible leakage, a seal washout of 20 cycles at twice rated flow at 60 degrees Celsius, and 72 hours of temperature cycling.

Note that 0.02 to 0.07 mL is a volume per mating cycle, not a leak rate per minute and not a bubble-test sensitivity. That scale difference determines the correct test method, converted in Section 7.

2. Get the flow right first: 13.4 g/s mislabelled as 13.4 L/min, a 16.64 times error

Volumetric flow through a cold plate follows from heat load, specific heat, density and supply-to-return temperature difference:

Take one GPU cold plate at 700 W, liquid-cooled fraction 80%, temperature difference 10 K, and 25 percent propylene glycol (cp = 3800 J/(kg*K), rho = 1036 kg/m3, consistent with our 2026-10-26 page):

Pure water as a cross-check (cp = 4180 J/(kg*K), rho = 998 kg/m3):

★ The number 13.4 is mass flow in g/s on the water basis. Labelling it 13.4 L/min treats 13.4 grams per second as 13.4 litres per minute. Back substitution:

A factor of 16.64. Sizing against 13.4 L/min skips the UQD02 band (below 4 L/min) and jumps to UQD06 (9 to 14 L/min), a five-fold jump in connector size. This is the same class of error recorded on 2026-10-26: heat transfer is computed with mass flow, pump and pipe sizing with volumetric flow, and mixing the two oversizes the whole loop.

3. Cv is not a lookup label: SI reduction of Q = Cv * sqrt(dP) and a self-consistency check

Cv is defined as US gallons per minute at 1 psi pressure drop. In SI:

Inverting, pressure drop at rated flow follows from rated flow and minimum Cv. For UQD02:

Applying the same method to all four sizes, with an effective-bore cross-check: assume a discharge coefficient Cd = 0.70, derive v = sqrt(2*dP/rho) and A = Q/(Cd*v), then compare the resulting diameter with the nominal bore.

Substituting the Cv definition forward gives Q_from_Cv matching rated flow to 0.00 percent for all four sizes (UQD02: 0.30 * 6.30901964e-5 * sqrt(3.996) * 60000 = 2.270 L/min). That confirms the minimum Cv and rated flow used here belong to the same Q = Cv * sqrt(dP) definition rather than being two unrelated table entries.

Effective bore falls 4.1, 8.1, 12.4 and 16.2 percent below nominal bore, with the gap growing with size. The reason is internal contraction and turning in the valve path, while Cd = 0.70 is only a uniform assumption; measured per-size Cd values would close the gap. Disclosed honestly: effective bore is an order-of-magnitude check, not a replacement for the vendor pressure-drop curve.

★ Version discrepancy: a second-hand summary labelled v2.0.1 gives minimum Cv as 0.25, 1.00, 1.60 and 2.40, which is 17 to 38 percent below the table above. At the same actual flow, the lower Cv set yields 5.6082 kPa for UQD02, 0.3505 kPa for UQD04, 0.1369 kPa for UQD06 and 0.0609 kPa for UQD08. Both sets are listed here; formal selection must use the primary OCP specification rather than a secondary table.

★ One more easily missed distinction: Cv is not the same in both flow directions through one coupling. The Parker UQDB brochure lists Socket-Plug and Plug-Socket directions separately as 0.30/0.33, 1.10/1.22, 2.25/2.60 and 3.90/4.09; UQDB06 differs by 15.6 percent, the largest of the four. Pressure-drop checks should use the smaller direction or resistance is underestimated. Amphenol measured Cv values are 0.4, 1.32, 2.11 and 3.83, with UQDB06 at 2.11 slightly below the v2.0.0 minimum of 2.20 - vendor measurements and specification minima are different quantities and must not be mixed.

4. Real duty point: pressure drop and pump power at 0.8535 L/min differ by 144 times

Inserting the flow from Section 2 into Q = Cv * sqrt(dP), with pump efficiency 0.40:

Smallest versus largest size gives a 144 times pressure-drop ratio, yet both absolute values are tiny: at most 0.14 W per coupling. At a branch flow of 0.85 L/min, pressure drop is simply not the controlling selection criterion; enlarging the connector to reduce it buys no engineering benefit.

The mislabelled 13.4 L/min case shows the cost of a broken unit chain: forced through UQD02 it implies 139.2 psi (960.0 kPa) and 535.99 W of pump power per coupling, roughly 38.5 kW across 72 couplings. That absurdity is a useful reductio.

5. The real cost of oversizing: mating force and effective pressurized area

The UQDB blind-mate specification gives a linear force law:

At the maximum operating pressure of 6.89 bar:

A widely used blind-mate reliability threshold is about 40 N; above it, a sliding server tray tends to reach only partial engagement. At 6.89 bar the four sizes reach 1.97, 3.28, 5.02 and 7.51 times that threshold:

★ Conclusion: even the smallest UQD02 exceeds the reliable blind-mate range when mated under pressure. The correct answer is to depressurize before mating, or to provide mechanical assist and floating compliance in the rack mechanism. Whether a coupling can be hot-mated under pressure is commonly treated as a coupling property; it is in fact a property of the rack mechanism.

Back-calculating effective pressurized poppet area from k

The pressure term arises from differential pressure acting on the projected poppet area:

As equivalent circular diameters these are 6.65, 11.07, 14.30 and 17.48 mm, all larger than the nominal bores of 3.2, 6.4, 9.5 and 12.7 mm. This is not an arithmetic slip: a spring poppet has shoulders, so the pressurized area includes the stem and seat annulus. The ratio to nominal bore declines with size (2.08, 1.73, 1.51, 1.38), indicating relatively more compact poppet geometry in larger sizes. The check is useful for verifying the order of magnitude of a vendor-supplied k.

Vendor data cross-check: k is not a paper coefficient

The Parker UQDB brochure gives mating force at zero pressure and at line pressure, so k can be derived independently. Take 1 lbf = 4.44822 N, 100 psi = 6.89 bar, 65 psi = 4.482 bar:

All four sizes fall within -7.0 to +7.5 percent. The linear F = F0 + k*p model matches measured vendor curves, confirming that k is a physical consequence of the pressurized poppet area rather than a fitted empirical constant, and that the 78.9 to 300.4 N computed in Section 5 at 6.89 bar has measured support. Parker zero-pressure values (35.6 to 102.3 N) sit below the OCP maxima (55 to 135 N) because of product design margin; vendor values must not be used to dispute the specification limits.

6. Physical precondition for 0.02 to 0.07 mL: the poppet must close before the seal separates

A dry-break coupling requires the poppet to close before the face seal parts. If that sequence fails, 6.89 bar acts directly on an open 6.4 mm bore. Bernoulli gives the jet velocity:

A poppet closing just 0.1 s late releases 3911 times the permitted fluid loss. That is why the specification treats loss per cycle as a hard limit: it measures poppet timing, not O-ring quality. Production acceptance should weigh collected fluid across real mate cycles rather than relying only on a hydrostatic hold test.

7. What actually enters the loop is air, not the lost liquid

When a dry-break coupling parts, the small volume trapped in the valve chamber is replaced by air. Using the UQD04 effective area of 96.2 mm2 and a chamber length of 8 mm:

Under that geometric assumption, chamber volume is 25.7 times the per-cycle loss limit, and air ingestion is governed by valve chamber geometry, not by the leak figure. But treating the effective pressurized area as the full chamber cross-section is an upper-bound estimate, and it must be tightened against measured data: the Parker UQDB brochure lists Spillage and Air Inclusion as two separate indicators, 0.004 / 0.080 mL for UQDB04 and 0.005 / 0.350 mL for UQDB08.

★ Self-correction: 0.770 mL corresponds to an equivalent chamber length of about 8 mm, while back-calculating from Parker measured 0.080 mL gives only 0.8316 mm - the original geometric estimate was roughly ten times too high. The direction of the conclusion is unchanged (air ingestion always exceeds fluid loss, and the gap grows with size: 20 times for UQDB04, 70 times for UQDB08), but the ratio must be tightened to the 2.67 to 70 band rather than treating 25.7 as a measured value. This is the only magnitude judgement on this page revised by external data; all other calculations are unaffected. Optimizing the sealing face to shave millilitres still does not reduce air intake.

As for the liquid loss itself: 72 servers, two couplings each, 12 mating cycles per year, at the UQD04 limit of 0.030 mL:

Against a 200 L loop this is 0.0259 percent, so topping up is negligible. The real risk is dissolved oxygen and rising conductivity from repeated air intake: our 2026-10-26 page gives galvanic current I = dE * kappa * A / L, where raising conductivity from 5 to 50 µS/cm raises aluminium consumption from 0.22 to 2.21 g per year, a tenfold increase. Maintenance thresholds should therefore watch conductivity and pH, not the make-up bottle level.

Leak detection must match the scale

Convert dry-break loss into the Pa*m3/s used by helium mass spectrometry. At atmospheric pressure, 1 Pa*m3/s corresponds to 1/101325 m3/s.

One dry-break event corresponds to 3.04e-3 Pa*m3/s, which is 304 times above the 1e-5 Pa*m3/s sensitivity typical of helium leak testing. It is a macroscopic droplet event, not molecular leakage, so helium testing is the wrong instrument; correct practice is gravimetric or collected-fluid measurement across mate cycles.

8. 316L body and EPDM seal: materials and a compression criterion

The specification restricts plug tip, shaft and guide area to SUS 303, 304, 304L, 316 or 316L. Brass bodies are unsuitable for deionized water: zinc leaches into the fluid, raising conductivity and forming a galvanic couple with copper cold plates and stainless mains. The advantage of 316L over 304 lies in chloride pitting resistance; the PREN breakdown on our 2026-11-02 page shows the gap comes mainly from molybdenum (304 about 19.5, 316L about 24.9). Trace chlorides from glycol degradation can concentrate 10 to 100 times inside a flange crevice, which is the quantitative reason to prefer 316L.

On the seal side, a reproducible compression criterion. Take an EPDM O-ring of 2.0 mm cord diameter in a 1.5 mm groove, with compression set estimated at 25 percent:

With only 18 percent initial compression (groove depth 1.64 mm):

Using 15 percent as the minimum sealing compression, the required initial compression follows:

★ Criterion: initial compression must be at least 20 percent to survive 25 percent compression set; practice takes 20 to 30 percent, and the 25 percent used here leaves five points of margin. If residual compression falls below 15 percent after 5000 cycles, fluid loss will exceed the limit before cycle life is exhausted. The O-ring is not a consumable but a component with a calculable life.

9. Common mistakes

1. Mistake: using mass flow in g/s directly as volumetric flow in L/min.
   Consequence: a 560 W cold plate reads 13.4 L/min instead of 0.85 L/min, a five-fold size jump, with mating force, envelope and cost all wrong.
   Fix: compute m = P/(cp*dT) first, then divide by density, checking units at each step.
2. Mistake: oversizing purely to cut pressure drop.
   Consequence: at 0.85 L/min the 02 and 08 sizes differ by only 3.867 kPa, while mating force at 6.89 bar rises from 78.9 N to 300.4 N.
   Fix: confirm pressure drop fits the pump head budget, then check mating force and the blind-mate mechanism.
3. Mistake: assuming a coupling can always be hot-mated under pressure.
   Consequence: even the smallest size reaches 78.9 N under pressure, 1.97 times the 40 N blind-mate threshold, inviting partial tray engagement.
   Fix: depressurize before mating, or provide mechanical assist and floating compliance in the rack.
4. Mistake: accepting dry-break performance with a helium leak detector.
   Consequence: a dry-break event is 3.04e-3 Pa*m3/s, 304 times above instrument sensitivity, so timing defects go undetected.
   Fix: use collected-fluid weighing across real cycles plus the hydrostatic hold per specification.
5. Mistake: watching make-up volume instead of conductivity.
   Consequence: annual loss is only 0.0259 percent of a 200 L loop, yet each disconnect admits air (vendor-measured 0.080 mL for UQDB04), and dissolved oxygen drives conductivity and galvanic corrosion.
   Fix: set maintenance thresholds on conductivity and pH together.

Basis statement: Numbers here fall into three classes. (1) Specification text (OCP UQD v2.0.0 rated flows, minimum Cv, fluid loss, mating force and its pressure coefficient, pressure and temperature limits). (2) Derived quantities computed from that text (pressure drop, effective bore, effective pressurized area, pump power, jet volume, chamber volume, compression ratios, leak-rate equivalent), each reproducible line by line from the parameters shown. (3) Engineering assumptions (Cd = 0.70, pump efficiency 0.40, 25 percent PG cp and rho, 25 percent compression set, 15 percent minimum sealing compression, 40 N blind-mate threshold, 72 servers at 12 cycles per year), all flagged in the text and requiring recomputation for other projects. Secondary sources disagree on minimum Cv by 17 to 38 percent; both sets are disclosed in Section 3, and the primary OCP specification governs formal selection.

One-line summary: 0.02 to 0.07 mL measures poppet timing rather than seal quality; separate g/s from L/min first, then invert Q = Cv * sqrt(dP) and F = F0 + k*p, and you find that at this flow pressure drop never controls selection - mating force and air ingestion do.

Frequently asked questions

What UQD size does a 700 W GPU cold plate actually need?

With 80 percent liquid cooling, 10 K temperature difference and 25 percent propylene glycol, mass flow is 14.7368 g/s and volumetric flow 0.8535 L/min, well below the 2.27 L/min UQD02 rating with 2.66 times margin. The widely copied 13.4 L/min is a mislabelled 13.4 g/s; back substitution gives 0.8040 L/min, a 16.64 times difference.

Can a UQD be blind-mated at 6.89 bar line pressure?

Using F = F0 + k*p from the specification, mating forces at 6.89 bar are 78.9, 131.3, 200.7 and 300.4 N, against a blind-mate reliability threshold of about 40 N; even the smallest size exceeds it 1.97 times. Depressurize before mating, or let the rack mechanism provide mechanical assist.

At 0.03 mL per cycle, how much coolant is lost in a year?

72 servers, two couplings each, 12 cycles per year gives 1728 cycles and 51.84 mL, only 0.0259 percent of a 200 L loop. But the valve chamber at 96.2 mm2 times 8 mm gives a geometric upper estimate of about 0.770 mL, far above the loss figure; vendor measurement gives 0.080 mL of air inclusion for UQDB04, see Section 7. Air ingestion and the resulting conductivity rise are the real maintenance control.