Cold Plate Pressure Drop: Pump and Flow Sizing
Short answer: Cold plate pressure drop is the hydraulic resistance your pump must overcome at the target flow rate, and it is usually the largest single loss in a liquid cooling loop. For a typical 1U–2U cold plate at 1–2 L/min, expect roughly 5–25 kPa (0.7–3.6 psi) per plate, with microchannel designs at the high end. Size the pump so its operating point sits at the intersection of the pump curve and the system resistance curve, then add 20–30% head margin. BQUQ quotes custom cold plates and heat sinks in 12 working hours.
Why pressure drop decides your whole loop
Engineers usually start liquid cooling projects by picking a cold plate for thermal resistance, then discover the pump cannot deliver the flow they assumed. Pressure drop and heat transfer are coupled: flow rate sets both the convective heat transfer coefficient and the hydraulic loss, and the loss rises faster than the benefit.
In a closed loop, the pump produces head (pressure) and the loop consumes it. Every component — cold plate, quick disconnects, tubing, fittings, radiator, reservoir — contributes. The pump's flow rate is not a fixed number; it is whatever flow the pump curve and the system resistance curve agree on. If you design around a datasheet "max flow" figure, you will almost always be disappointed.
The practical consequence: pressure drop is not a detail you check at the end. It is a design input that determines pump selection, tubing diameter, coolant choice, and even whether liquid cooling is viable versus a heat pipe or vapor chamber approach.
What actually creates pressure drop in a cold plate?
Pressure drop in a cold plate comes from four mechanisms, and knowing which dominates tells you where to spend design effort.
Friction along channels
Viscous shear against channel walls produces the baseline loss. For laminar flow in a rectangular channel, ΔP scales linearly with velocity and inversely with the square of the hydraulic diameter. Halving channel width roughly quadruples the loss at the same flow rate.
Minor losses at inlet and outlet
Sudden contraction at the inlet manifold and sudden expansion at the outlet create irreversible losses. These are captured by a loss coefficient K, typically 0.5–1.5 each, and scale with velocity squared. In short cold plates with wide channels, minor losses can exceed friction losses.
Manifold and distribution losses
A poorly designed manifold forces uneven flow across parallel channels. The channels that receive less flow run hotter, and the whole plate must be oversized to compensate. Good manifold design costs a little pressure drop but buys uniform temperature — usually the right trade.
Bend, fitting, and tubing losses
Outside the plate itself, elbows, quick disconnects, and narrow tubing add up. A single pair of dry-break QDs can add 5–15 kPa, comparable to the cold plate itself.
Reading a pressure drop curve
Cold plate vendors publish ΔP versus flow rate, usually as a power law: ΔP = a·Q^b, where b is roughly 1 for fully laminar flow and approaches 2 for turbulent flow. Most electronics cold plates operate in the transitional range, so b lands between 1.3 and 1.8.
| Flow rate (L/min) | Typical 1U cold plate ΔP (kPa) | Typical microchannel ΔP (kPa) | Typical large-format plate ΔP (kPa) |
|---|---|---|---|
| 0.5 | 3–6 | 10–18 | 1–3 |
| 1.0 | 8–14 | 25–45 | 3–7 |
| 1.5 | 15–24 | 45–80 | 6–12 |
| 2.0 | 24–38 | 70–120 | 10–18 |
Values are indicative for water–glycol at 25 °C and will shift with coolant viscosity, temperature, and internal geometry. Always request a measured curve for the specific part.
Why the exponent matters for pump sizing
Because ΔP rises faster than flow, doubling flow more than doubles the required head. If you need 30% more flow to hit a thermal target, expect roughly 70–100% more pressure drop. That nonlinearity is why "just turn up the pump" fails: the pump curve is relatively flat near its maximum flow, so head collapses exactly when you need it.
Building the system resistance curve
The full loop resistance is the sum of every component's ΔP at each candidate flow rate:
ΔP_system(Q) = ΔP_coldplate + ΔP_QDs + ΔP_tubing + ΔP_radiator + ΔP_fittings + ΔP_reservoir
For tubing, use the Darcy–Weisbach equation with the appropriate friction factor. For fittings, use K-factors. For the radiator, use the vendor curve. Then plot the sum against Q.
| Loop element | Typical share of total ΔP | Notes |
|---|---|---|
| Cold plate(s) | 40–65% | Dominant in single-plate loops |
| Quick disconnects | 10–25% | Highly model-dependent |
| Tubing and hose | 8–20% | Strongly dependent on ID |
| Radiator / heat exchanger | 10–25% | Vendor curve required |
| Fittings, bends, reservoir | 5–15% | Often underestimated |
The takeaway: the cold plate is usually the biggest single contributor, but the "small stuff" can easily total 30–40%. Ignoring it leads to undersized pumps and thermal throttling.
Matching the pump to the loop
Once you have the system curve, overlay it on candidate pump curves. The intersection is your operating point.
Step-by-step sizing procedure
1. Define the required flow rate from the thermal budget, not from the pump datasheet.
2. Compute ΔP for each loop component at that flow, and at 0.5× and 1.5× that flow.
3. Plot the system resistance curve.
4. Overlay pump curves (P-Q curves) for candidate pumps.
5. Find the intersection. Verify it is at or below the target flow's required head.
6. Add 20–30% head margin for fouling, coolant aging, and manufacturing tolerance.
7. Check the pump's maximum pressure rating and the cold plate's burst pressure.
Worked example
Suppose a 300 W IGBT module needs a 6 °C case-to-coolant rise. A cold plate with 0.02 °C/W thermal resistance at 1.5 L/min would give 6 °C at 300 W — so 1.5 L/min is the target. At that flow, the cold plate drops 20 kPa, QDs drop 10 kPa, tubing drops 6 kPa, radiator drops 8 kPa, fittings drop 4 kPa. Total: 48 kPa. Add 25% margin: 60 kPa at 1.5 L/min. Select a pump whose curve passes through or above 60 kPa at 1.5 L/min.
That single number — 60 kPa at 1.5 L/min — is what you take to pump selection. Everything else is detail.
Design choices that reduce pressure drop
If the required pump is too large, too loud, or too expensive, reduce loop resistance rather than buying more head.
Widen the channels
Going from 0.5 mm to 1.0 mm channel width can cut ΔP by 60–75% at the same flow, at the cost of some thermal performance. If you have thermal margin, this is the cheapest fix.
Shorten the flow path
A serpentine path 300 mm long has far more friction than a 100 mm path with a parallel manifold. Parallel flow paths reduce ΔP but require careful manifold design to keep flow uniform.
Use larger tubing and fewer fittings
Tubing ΔP scales with the inverse fourth to fifth power of internal diameter in turbulent flow. Going from 6 mm to 8 mm ID can cut tubing losses by more than half.
Choose low-loss quick disconnects
QD loss varies enormously between models. A low-loss QD pair might add 3 kPa where a standard pair adds 15 kPa. For high-flow loops, this is often the single biggest easy win.
Consider coolant properties
Water–glycol mixtures have 20–40% higher viscosity than water at the same temperature, which increases ΔP proportionally in laminar flow. If freeze protection is not required, plain water with biocide and corrosion inhibitor is the lower-loss choice.
Where manufacturing tolerance enters
Pressure drop is not just a design number — it is a manufactured number. Channel width tolerance, surface roughness, and manifold geometry all shift the real ΔP away from the model. A cold plate machined to ±0.005 mm on channel dimensions behaves predictably; one with loose tolerances may run 10–20% off the curve.
At BQUQ, cold plates and liquid-cooling heat sinks are produced on CNC machining lines holding ±0.005 mm, alongside metal stamping, custom springs, and heat sink production across four lines in one Dongguan factory. That tolerance control matters most for microchannel and skived-fin plates, where a few tens of microns of channel variation changes both thermal resistance and pressure drop. For designers comparing architectures, our overview of custom heat sink manufacturing and CNC machined heat sinks covers the trade-offs between skived, bonded, and machined constructions. If your loop uses extruded profiles, extruded heat sinks can be adapted into liquid-cooled plates with machined manifolds.
For teams modeling the thermal side before committing to hardware, heat sink thermal simulation tools and the thermal resistance network approach are useful companions to the hydraulic analysis here. And if your application is pushing past what a single-phase cold plate can do, vapor chamber design is the next step to evaluate.
Frequently Asked Questions
Q: What is a normal pressure drop for a cold plate?
A: For a 1U–2U single-phase cold plate at 1–2 L/min with water–glycol, 5–25 kPa (0.7–3.6 psi) is typical. Microchannel plates run higher, often 40–100 kPa at the same flow. Large-format plates with wide channels can be under 5 kPa. Always use the vendor's measured curve for your specific geometry and coolant rather than a rule of thumb.
Q: How do I calculate the required pump head?
A: Sum the pressure drop of every loop component at your target flow rate, then add 20–30% margin. The result is the head the pump must produce at that flow. Plot this against candidate pump curves and pick the pump whose curve passes through or above that point. Do not size from the pump's maximum flow rating, which occurs at zero head.
Q: Does higher flow always improve cooling?
A: No. Thermal resistance falls with flow but with diminishing returns, while pressure drop rises faster than flow. Past a certain point, extra flow buys very little temperature improvement but requires disproportionately more pump power and noise. Find the knee of the thermal resistance curve and design near it, not beyond it.
Q: How does coolant choice affect pressure drop?
A: Viscosity drives it. Water–glycol mixtures are 20–40% more viscous than water at the same temperature, raising laminar-flow pressure drop proportionally. If freeze protection is unnecessary, water with corrosion inhibitor and biocide gives the lowest pressure drop. Always confirm material compatibility with your cold plate's wetted metals.
Q: Can BQUQ help with cold plate design and quoting?
A: Yes. BQUQ runs CNC machining, metal stamping, custom springs, and heat sink production across four lines in one ISO9001 factory in Dongguan. We hold ±0.005 mm on machined features, support flexible MOQ, and return quotes within 12 working hours. Send your drawing, flow target, and coolant specification to sc@bquq.com.
Related Resources
- About BQUQ and our Dongguan manufacturing footprint: /about/
- Heat sink product families, including liquid-cooling plates: /heat-sinks/
- Extruded heat sink profiles for custom loops: /extruded-heat-sinks/
- CNC machined heat sinks and cold plates: /cnc-machined-heat-sinks/
- Industry trends in electronics cooling: /industry-dynamics/
- Technical articles and engineering guides: /bquq-blog/
- Frequently asked questions: /faq/
- Case studies: /case/
- Contact the engineering team: /contact/
Authored by the BQUQ Engineering Team. BQUQ (Dongguan) runs CNC machining (±0.005 mm), metal stamping, custom springs, and heat sink production in one ISO9001 factory. Source-direct from Dongguan, China — quote in 12 hours: sc@bquq.com | WhatsApp +86 13713157787 | www.bquq.com
