Immersion cooling fluid comparison using a single-phase immersion/liquid technique for data centers

Overview

The study asks how different immersion fluids perform in a hybrid server. Fluids were ranked with the OCP Immersion Cooling Sub-Project Figure of Merit for natural convection (FOM1), component temperatures were measured, and a CFD model was validated for wider fluid screening; the CFD model matched experiments with a maximum temperature deviation of 10.7% (NVMe average temperature). SL 3568 properties: density 838/812/773 kg/m³ and kinematic viscosity 406/43/7.6 cSt at 0/40/100 °C, with thermal conductivity 0.142-0.134 W/m·K.

Roughly doubling FOM1 (BioLife 4 versus SL 3568) reduces the Prandtl number by about a factor of ten. A near doubling of FOM1 lowered average fluid temperature by approximately 7.7 °C. With roughly doubled FOM1, immersed RAMs, NVMes and SSDs ran about 6.7, 6.4 and 5.9 °C cooler respectively. Higher-FOM1 fluids cool immersed components better.

Unexpectedly, they make cold-plated CPUs hotter: CPU temperatures were highest with high-FOM1 fluids, attributed to dielectric fluid infiltrating the TIM between CPU and water block and reducing TIM viscosity by at least 94.4%. This is a compatibility risk for hybrid immersion/direct liquid cooling designs.

Limits: a single hybrid server and four commercial fluids; results are for this TIM (KP98) and server and must be re-verified for other TIM and fluid pairings. It shows a trade-off between cooling performance and TIM compatibility.

Key points

1. Roughly doubling FOM1 (BioLife 4 vs SL 3568) reduces the Prandtl number by about a factor of ten.
2. A near doubling of FOM1 lowered average fluid temperature by approximately 7.7 °C.
3. With roughly doubled FOM1, immersed RAMs, NVMes and SSDs ran about 6.7, 6.4 and 5.9 °C cooler respectively.
4. CPU temperatures were highest with high-FOM1 fluids, attributed to dielectric fluid infiltrating the TIM between CPU and water block and reducing TIM viscosity by at least 94.4%.
5. The CFD model matched experiments with a maximum temperature deviation of 10.7% (NVMe average temperature).
6. SL 3568 properties: density 838/812/773 kg/m³ and kinematic viscosity 406/43/7.6 cSt at 0/40/100 °C, thermal conductivity 0.142-0.134 W/m·K.

Using it in design

• In hybrid immersion/direct liquid cooling design, rank dielectric fluids by FOM1, but remember a high-FOM1 fluid can infiltrate the TIM of cold-plated CPUs and raise their temperature.
• Add fluid and TIM compatibility (viscosity drop of 94.4%) to design verification; in hybrid systems select the fluid together with a CPU cooling test.
• For immersed components (RAM, NVMe, SSD), a higher-FOM1 fluid gave 6-7 °C cooler operation; compare component-level temperature targets with that gain.

Who should read it

Cooling engineers selecting dielectric fluids for single-phase immersion and hybrid direct liquid cooling.

Document contents

1. 1. Introduction
2. 3. Fluids comparison experimental test (3.1 Experimental setup and procedure, 3.2 Results and discussion: FOM1 criteria evaluation)
3. 4. Numerical methodology (4.1 Model description, 4.2 Grid independence analysis, 4.3 Results and discussion: water and fluid behavior, IT components behavior)
4. 5. Conclusion

Related standards

• OCP Immersion Cooling Sub-Project (FOM1)

Read the document

Publication details

Authors

Ayoub Ruichek; Mohamad Hnayno; Ali Chehade; Farouk Fardoun; Tala Moussa; Guillaume Polidori; Chadi Maalouf

Publisher

Applied Thermal Engineering (Elsevier BV)

Year

2025

Volume

280

Pages

128355

Document language

English

Source type

Scientific

Access

Open licence — downloadable

DOI

10.1016/j.applthermaleng.2025.128355

Licence

CC BY 4.0 — share with attribution

Licence evidence

hal.science

Citation