Thermal resistance (Rth) of a liquid cold plate is the single most important performance metric of any liquid cooling system — it determines how efficiently heat is transferred from the device junction to the coolant. Rth is expressed in °C/W: a cold plate with an Rth of 0.04 °C/W will see its coolant temperature rise by 4 °C for every 100 W of heat dissipated. For a 2400 W GPU, that translates into a 96 °C rise from coolant to junction — which is exactly why optimizing Rth is mission-critical for high-power AI computing.
What Is Liquid Cold Plate Thermal Resistance?
Liquid cold plate thermal resistance (Rth) is defined as the temperature difference between the heat source junction (Tj) and the bulk coolant temperature (Tcoolant), divided by the total heat dissipated (Q):
Rth = (Tj − T_coolant) / Q [°C/W]
The total cold plate Rth is the sum of three thermal resistances in series:
• Thermal interface material resistance (R_TIM) — the TIM between the device and the cold plate base.
• Base conduction resistance (R_base) — heat spreading through the cold plate base material.
• Convective resistance (R_conv) — heat transfer between the coolant and the fins (the dominant term).
How to Calculate Cold Plate Thermal Resistance
Convective resistance (R_conv) accounts for the majority of total Rth. It is calculated as:
R_conv = 1 / (h × A_wetted)
where h is the heat transfer coefficient (W/m²·K) and A_wetted is the total wetted surface area of the channel structure.
For turbulent flow inside rectangular microchannels, the heat transfer coefficient is calculated using the Dittus–Boelter correlation:
Nu = 0.023 × Re^0.8 × Pr^0.4
h = Nu × k_fluid / D_h
where Re is the Reynolds number, Pr is the Prandtl number of the coolant, k_fluid is the coolant’s thermal conductivity, and D_h is the hydraulic diameter of the channel.
Worked Example: Rth Calculation for a GB300 Cold Plate
A copper microchannel cold plate for the GB300 GPU (2400 W, 152 × 102 mm footprint):
• Channel width: 1.0 mm, channel height: 5 mm, fin pitch: 1.6 mm
• Channel count: 70, length per channel: 90 mm
• Coolant: 40% ethylene glycol / water, total flow rate: 3.5 L/min
Calculations
• A_wetted = 70 × 2 × (1.0 + 5.0) × 10⁻³ × 90 × 10⁻³ = 0.0756 m²
• D_h = 2 × (1.0 × 5.0) / (1.0 + 5.0) mm = 1.67 mm
• Re ≈ 1,450 at 3.5 L/min across 70 channels (transitional / turbulent)
• h ≈ 18,500 W/m²·K
• R_conv per channel = 1 / (18,500 × 0.0756) = 0.0007 °C/W; aggregate R_conv ≈ 0.026 °C/W
The measured Rth for this configuration is 0.027 °C/W (including R_base and R_TIM with indium foil) — within 4% of the analytical prediction.
How to Measure Cold Plate Thermal Resistance
Accurate Rth measurement requires a calibrated Thermal Test Vehicle (TTV). The procedure:
1. Mount the TTV on the cold plate using a production-grade TIM, torqued to spec.
2. Set coolant flow rate to the target value (measured with a Coriolis or turbine flow meter, ±1% accuracy).
3. Use a recirculating chiller to hold coolant inlet temperature at 40 °C ± 0.5 °C.
4. Apply the target thermal load (e.g., 2400 W) through the TTV’s embedded heaters.
5. Wait for steady state: ΔTj < 0.1 °C over 5 minutes.
6. Record Tj (TTV thermocouples), Tin, Tout, and Q (power accuracy ±0.5%).
7. Calculate: Rth = (Tj − Tin) / Q
Note: A common measurement mistake is testing Rth at partial load (e.g., 500 W instead of full TDP). Rth is not constant — it decreases slightly at high flow due to the laminar-to-turbulent transition. Always measure and specify Rth under realistic operating conditions.
How to Optimize Cold Plate Thermal Resistance
Based on sensitivity analysis across a wide range of cold plate CFD studies, four levers consistently dominate Rth optimization:
1. Increase Flow Rate (Highest Impact at Low Flow)
Increasing flow from 1.0 L/min to 2.0 L/min typically reduces Rth by 25–35%, driven by a higher Reynolds number and a shift from laminar to turbulent flow. Above 3.0 L/min, the Rth reduction per doubling drops below 10% while pressure drop scales quadratically. For most GPU cold plate applications, the sweet spot is 2.5–4.0 L/min.
2. Reduce Channel Hydraulic Diameter (Microchannel Design)
Reducing channel width from 2.0 mm (mini-channel) to 1.0 mm (microchannel) increases h by roughly 40% through a higher surface-area-to-volume ratio. The trade-off: pressure drop rises 3–4×. CFD optimization is the standard way to find the Pareto-optimal channel geometry for each application’s flow budget.
3. Switch from an Aluminum to a Copper Base
Replacing an aluminum base (k = 167 W/m·K) with copper (k = 400 W/m·K) reduces R_base by about 58%. The impact is greatest for high-heat-flux applications (> 50 W/cm²) where base-spreading resistance dominates. For diffuse heat sources (< 20 W/cm²), the impact is marginal.
4. Upgrade TIM from Paste to Indium Foil
Switching from high-performance thermal paste (k = 8–12 W/m·K) to indium foil (k = 82 W/m·K, 100 μm thick) reduces R_TIM from 0.012–0.020 °C/W to 0.002–0.004 °C/W. At 2400 W, that saves 24–48 °C of thermal budget — the equivalent of moving from a mini-channel to a microchannel design. Indium foil is made from 99.995% indium ingot; raw material cost is low and price volatility is significantly lower than for many competing materials.
Six reasons indium foil is the preferred thermal interface:
8. Excellent thermal conductivity
9. Good flexibility and malleability
10. High ductility
11. Strong compatibility with semiconductor packaging
12. Low-cost TIM solution
13. Broad and growing application landscape
FAQ: Liquid Cold Plate Thermal Resistance
What is the ideal thermal resistance for a GPU liquid cold plate?
The ideal Rth depends on GPU TDP: NVIDIA H200 (1000 W) targets Rth ≤ 0.050 °C/W; GB200 (1800 W) targets Rth ≤ 0.040 °C/W; GB300 (2400 W) targets Rth ≤ 0.030 °C/W. These targets keep junction temperatures below 85 °C at a standard 40–45 °C coolant inlet temperature.
How does coolant flow rate affect thermal resistance?
Higher flow reduces Rth by increasing the convective heat transfer coefficient — but the relationship is non-linear. Going from 1 to 2 L/min typically cuts Rth by 25–35%, while going from 3 to 6 L/min only yields an 8–12% reduction. Beyond 3–4 L/min, diminishing returns set in, meaning channel-geometry optimization delivers more Rth-per-unit-pressure-drop than simply pushing more flow.
What is the difference between junction-to-case and junction-to-coolant thermal resistance?
Junction-to-case resistance (Rjc) is a device-level parameter defined by the semiconductor manufacturer, representing heat flow from the die to the bottom of the package. Junction-to-coolant resistance (Rth, total) includes Rjc plus the TIM resistance, cold plate base resistance, and convective resistance. Cold plate Rth specifications typically refer only to the cold plate itself (case-to-coolant) and exclude Rjc. When comparing vendor specs, always confirm which definition is being used.
-
Written by
CoolingThermal Engineering TeamCoolingThermal is an automation equipment manufacturer based in Kunshan, China, specializing in heat pipe and vapor chamber production equipment since 2017. Our engineering team designs, builds, and commissions complete production lines covering forming, degassing, welding, testing, and assembly processes. The technical content on this blog is written by the same team that develops the equipment — based on real production experience, not secondary research.
