How to Size a Coolant Pump for a Liquid-Cooling Loop

Liquid Loop Lab · Practical liquid-cooling engineering

Every pump selection comes down to two numbers: the flow rate your heat load demands, and the pressure your loop resists that flow with. Get those two right and the pump almost picks itself. Get either wrong and you end up with a loop that overheats at load, a pump screaming at the end of its curve, or both. This guide walks the whole method, with a worked example you can check against the free calculator.

Step 1 — Flow rate from the heat load

The physics is one equation:

Q = ṁ · c_p · ΔT

Heat load Q (watts) equals mass flow ṁ (kg/s) times the coolant's specific heat times the temperature rise you allow across the load. Rearranged, the mass flow you need is Q / (c_p · ΔT). Water's specific heat is about 4,180 J/kg·K; a 25% propylene-glycol mix is nearer 3,900, and 50% glycol drops to roughly 3,550 — one reason glycol loops need more flow for the same job.

Choosing ΔT is a design decision, not a constant. A small ΔT (3–5 K) keeps every device in the loop near the supply temperature, but demands high flow — and pumping power rises steeply, because pressure drop grows with roughly the square of flow. A large ΔT (8–12 K) cuts flow and pump size, but downstream devices drink warmer coolant and your facility-water approach temperatures get squeezed. Most CDU loops land at 5–10 K. Pick deliberately, and write the choice down — it drives everything below.

Step 2 — Budget the pressure drop

With flow known, add up everything that resists it, at that flow:

• Straight runs — Darcy–Weisbach: friction factor (Haaland is fine) times L/D times the dynamic pressure. Small tube is expensive: halving the diameter raises drop roughly 32× at the same flow.
• Fittings — each elbow, tee, and valve adds a loss coefficient K. A tidy ΣK for a simple loop is 5–15.
• Components — cold plates, filters, and heat exchangers. Get these from vendor ΔP-vs-flow curves, never guesses; they usually dominate the budget.
• Quick disconnects — the classic surprise. A pair of no-spill QDs can cost as much pressure as metres of tubing. Always pull their ΔP at your design flow.

Sum the pieces at design flow for the total loop drop. Then check the worst case: cold glycol on a winter start can be several times more viscous than at operating temperature, and laminar pressure drop scales directly with viscosity. If the pump can't push cold fluid, the loop can't start.

Step 3 — Match the pump curve, not the headline numbers

A pump doesn't deliver "its" flow or "its" pressure — it delivers the point where its curve crosses your loop's resistance curve. So plot (or tabulate) both: the pump's head vs flow, and your loop's ΔP rising with flow². The intersection is your operating point.

• Never select on maximum flow (that's at zero head) or maximum head (that's at zero flow). Both numbers are real; neither is achievable in your loop.
• Land the design point near the pump's best-efficiency region, typically the middle third of the curve — efficiency, noise, and bearing life all live there.
• Hold 10–20% flow margin for fouling, filter loading, and the ΔP you forgot. But don't oversize wildly: an oversized pump throttled back is wasted energy and heat added to your own coolant.

Worked example — a 10 kW loop

So the shopping spec reads: a pump whose curve passes through (or above) 28.8 L/min at 35 kPa, with best efficiency near that point — not "a 30 L/min pump" and not "a 100 kPa pump." Run your own numbers in the calculator; the same page is the worked example, live.

Five mistakes that bite first builds

1. Selecting on headline max-flow or max-head. See Step 3.
2. Ignoring cold-start viscosity. Size the check at the coldest, thickest coolant the pump will ever see.
3. Forgetting the quick disconnects. QDs are components with curves; budget them like cold plates.
4. No high-point bleed. Trapped air collects at the high point, starves a branch, and makes a liar of your flow meter. Design the bleed in.
5. Assuming parallel branches share flow fairly. They share it by resistance, not by fairness. Balance them on paper (the kit includes a worksheet), then verify with flow or ΔT per branch at bring-up.