본문 바로가기
Data Center Technology Lab Data Center Technology Lab

Coolant Types in the Data Center and What Each Sheet Assumes

A PG25 coolant sheet lists electrical conductivity above 2,000 micromho/cm. ASHRAE guidance for water in the same loop sets a range of 0.2 to 20 micromho/cm. On heat carried per liter of flow, the two fluids differ by 3.1 percent. The choice among water, glycol and dielectric fluid turns on what the coolant touches and what it does after it escapes, not on how much heat it holds.

Water and PG25 need nearly the same flow per kilowatt

The governing equation leaves little room between water and PG25. Heat carried equals mass flow × specific heat × temperature rise, written Q = m-dot × cp × ΔT. Solved for volume, the flow each kilowatt requires is V = Q ÷ (ρ × cp × ΔT), where ρ is density.

The assumptions are 1 kW of heat and a 10 K rise. Properties are read at 50 °C, the return side of a loop supplied at 40 °C. Dow’s DOWFROST LC 25 sheet gives PG25 a density of 1,016.8 kg/m³ and a specific heat of 3.94 kJ/kg·K at that temperature.

Substituted, 1 kW ÷ (1,016.8 kg/m³ × 3.94 kJ/kg·K × 10 K) equals 2.496 × 10⁻⁵ m³/s, or 1.50 L/min per kW. The result is calculated from typical properties. It is not a measurement.

The Open Compute Project’s OAI liquid cooling guideline names a 10 °C rise as the typical design target and states it as 1.5 LPM/kW, based on the properties of PG25-based coolants. The arithmetic reproduces that figure. OCP’s ratio is a property calculation, not a cold plate rating; it fixes flow and says nothing about chip temperature.

Source: Open Compute Project, “OAI System Liquid Cooling Guidelines” (March 2023 update); Dow, “DOWFROST LC Heat Transfer Fluid” technical data sheet, Form No. 180-01627-01-0523

Water at 50 °C, with a density of 988.03 kg/m³ and a specific heat of 4.1806 kJ/kg·K, needs 1.45 L/min per kW under the same assumptions. PG25 therefore needs 3.1 percent more flow than water.

Table 1. Coolant flow required per kilowattL/min per kW, 10 K rise

FluidProperties read atDensity (kg/m³)Specific heat (kJ/kg·K)Flow (L/min per kW)
Water50 °C988.034.18061.45
PG25 (DOWFROST LC 25)50 °C1,016.83.941.50
PG55 (DOWFROST LC 55)50 °C1,034.93.431.69
Hydrocarbon dielectric fluid (Shell S5 X)*15 °C density, 40 °C specific heat8062.2743.27

* Mixed property temperatures as published; flow is a lower bound at 40 °C

Flow = 1 kW ÷ (density × specific heat × 10 K); calculated from typical properties, not a cold plate rating

PG25 row matches the OCP OAI design target of 1.5 LPM/kW at a 10 °C rise

Source: Dow, “DOWFROST LC Heat Transfer Fluid” TDS, Form No. 180-01627-01-0523; Shell, “Shell Immersion Cooling Fluid S5 X” TDS v1.5 (2021); CRC Handbook of Chemistry and Physics, 84th Ed. (2004) via LibreTexts; Open Compute Project, “OAI System Liquid Cooling Guidelines” (2023)

A 3.1 percent flow penalty does not decide a coolant. Pump selection absorbs it. The reasons to pick one fluid over the other sit in the chemistry.

Source: CRC Handbook of Chemistry and Physics, 84th Edition (2004), as tabulated in LibreTexts Chemistry, “S2: Properties of Water”

The glycol in a data hall is there for biology, not frost

The OCP guideline for propylene glycol fluids ties the 25 percent figure to microbial control. It states that biocides are not necessary when glycol stays above 25 percent by volume, because the fluid is then bio-static, and it sets the allowed band at 24.5 to 29.5 percent.

Freeze protection comes with it, at -13 to -9 °C for PG25 in the same guideline, but a conditioned data hall rarely calls on it. ASHRAE guidance for water in this loop takes the other route and lists a biocide as required.

Source: Open Compute Project, “Guidelines for Using Propylene Glycol-Based Heat Transfer Fluids in Single-Phase Cold Plate-Based Liquid Cooled Racks”; ASHRAE TC 9.9, “Water-Cooled Servers: Common Designs, Components, and Processes” (2019)

Raising concentration past that floor costs viscosity. At 50 °C, Dow lists 1.27 mPa·s for PG25 and 2.95 mPa·s for PG55.

PG55 also needs 1.69 L/min per kW in the same arithmetic, because its specific heat falls to 3.43 kJ/kg·K. I read the 25 percent figure as a biological floor rather than a thermal choice. Going higher buys frost protection that matters only if the loop ships, sits in storage or runs below freezing.

Ethylene glycol appears in OCP’s OAI guideline beside deionized water and propylene glycol as one of the most common cold plate coolants. A Schneider Electric engineering blog notes that it performs somewhat better thermally but is far more toxic, and that server quick disconnects often leave a slightly wetted break during service. That service pattern, not heat transfer, keeps ethylene glycol out of most racks.

Source: Schneider Electric, “What coolant should you use for data center liquid cooling?” (August 14, 2026)

Fresh PG25 meets the ASHRAE water limits, and in service it may run five times higher

The OCP glycol guideline carries two quality tables that look alike and are not. Its Table 1, the typical properties of the fluid, sets chloride below 5 ppm, sulfate below 10 ppm and total hardness below 20 ppm.

Those three limits match the ASHRAE TCS (technology cooling system, the loop between the coolant distribution unit and the servers) water guidance line for line, as reproduced in the TC 9.9 water-cooled servers white paper.

The OCP document’s Table 4, the quality to maintain in operation, allows chloride below 25 ppm, sulfate below 25 ppm and hardness below 100 ppm.

The guideline does not label the tables this way, but I read Table 1 as an acceptance standard for delivered fluid and Table 4 as an operating limit. A buyer who writes Table 4 into a purchase specification accepts fluid carrying five times the chloride the guideline expects on delivery.

Source: Open Compute Project, “Guidelines for Using Propylene Glycol-Based Heat Transfer Fluids in Single-Phase Cold Plate-Based Liquid Cooled Racks”; ASHRAE TC 9.9, “Water-Cooled Servers: Common Designs, Components, and Processes” (2019)

Conductivity separates the two fluids further. ASHRAE sets 0.2 to 20 micromho/cm for TCS water, while Dow lists both PG25 and PG55 above 2,000 micromho/cm.

In PG25 the inhibitor package is the conductivity. A monitoring plan built for a deionized water loop, where rising conductivity flags contamination, would read a healthy PG25 loop as a failed one. OCP’s quarterly on-site checks are pH, glycol concentration by refractometer, and appearance. The same figure means a PG25 leak onto a board is a leak of a conductive liquid.

Source: Dow, “DOWFROST LC Heat Transfer Fluid” technical data sheet, Form No. 180-01627-01-0523

Dielectric fluid tolerates contact and needs more than twice the flow

Dielectric fluid is chosen when the coolant must touch the electronics, which in current practice means immersion. Shell’s Immersion Cooling Fluid S5 X sheet lists a breakdown voltage of 42 kV and a D.C. resistivity of 9.9 TΩ·m at 60 °C.

Its thermal figures are weaker. The sheet gives a specific heat of 2,274 J/kg·K at 40 °C and a density of 806 kg/m³ at 15 °C.

The same arithmetic gives 1 kW ÷ (806 kg/m³ × 2.274 kJ/kg·K × 10 K) = 5.456 × 10⁻⁵ m³/s, or 3.27 L/min per kW. That is 2.25 times the water figure.

Viscosity widens the gap. Shell lists a kinematic viscosity of 9.8 mm²/s at 40 °C, where Dow’s PG25 values work out to 1.545 mm²/s.

By that measure S5 X is 6.3 times more viscous than PG25. On a cold plate loop, those numbers buy nothing the water-based fluids lack. The case for a dielectric fluid is the case for removing the cold plate: the coolant reaches every component, and a leak becomes a cleanup rather than a short. Whether that trade suits a given hall is a question for engineering review, not a property comparison.

Fluorinated dielectric fluid, the family used in two-phase immersion, adds a supply question. 3M announced in December 2022 that it would exit PFAS manufacturing, fluorinated fluids included, by the end of 2025. ASHRAE TC 9.9 had already noted in 2019 that some engineered fluids showed significant material issues and that others carried global warming potential high enough to rule them out.

Source: Shell, “Shell Immersion Cooling Fluid S5 X” technical data sheet, v1.5 (2021); 3M, “3M to Exit PFAS Manufacturing by the End of 2025” (December 20, 2022); ASHRAE TC 9.9, “Water-Cooled Servers” (2019)

Conditions the sheets do not hold constant

The Shell sheet reports density at 15 °C and specific heat at 40 °C. Hydrocarbon density falls as temperature rises, so 3.27 L/min per kW is a lower bound for the fluid running at 40 °C.

Dow labels its tables typical properties, not specifications. Its summary table also rounds the 20 °C viscosity of PG25 to 2.8 mPa·s, where the full property table gives 2.84.

None of these sheets states a heat transfer coefficient, because that depends on the channel geometry of a particular cold plate or tank. Whether 3.1 percent more flow, or 6.3 times the viscosity, changes chip temperature needs the hardware vendor’s data and engineering review.

Source: Shell, “Shell Immersion Cooling Fluid S5 X” technical data sheet, v1.5 (2021); Dow, “DOWFROST LC Heat Transfer Fluid” technical data sheet, Form No. 180-01627-01-0523

Compare sheets at one temperature, then ask what the fluid touches

The sheets settle the thermal question in a few lines of arithmetic. The selection turns on chemistry and contact. Before comparing two coolant sheets, check these:

  • Density and specific heat read at the same temperature, near the loop’s actual supply or return
  • Flow per kW rebuilt from the sheet; OCP’s 1.5 LPM/kW reproduces, so a vendor ratio that does not deserves a question
  • Whether a quality table is an as-delivered specification or an in-service limit
  • A monitoring plan matched to the fluid: conductivity for deionized water, pH and concentration for PG25
  • What the fluid wets if it escapes, and, for fluorinated fluid, who manufactures it after 2025

On heat per liter, water and PG25 behave as one fluid. On chemistry they are two, and a dielectric fluid is a different system altogether.

Source: Open Compute Project, “OAI System Liquid Cooling Guidelines” (March 2023 update)

References

Open Compute Project, “OAI System Liquid Cooling Guidelines” / Open Compute Project / March 3, 2023 update / https://www.opencompute.org/documents/oai-system-liquid-cooling-guidelines-in-ocp-template-mar-3-2023-update-pdf

Open Compute Project, “Guidelines for Using Propylene Glycol-Based Heat Transfer Fluids in Single-Phase Cold Plate-Based Liquid Cooled Racks” / Open Compute Project / undated / https://www.opencompute.org/documents/guidelines-for-using-propylene-glycol-based-heat-transfer-fluids-in-single-phase-cold-plate-based-liquid-cooled-racks-final-pdf

ASHRAE TC 9.9, “Water-Cooled Servers: Common Designs, Components, and Processes” / ASHRAE / 2019 / https://www.ashrae.org/file%20library/technical%20resources/bookstore/whitepaper_tc099-watercooledservers.pdf

Dow, “DOWFROST LC Heat Transfer Fluid” technical data sheet, Form No. 180-01627-01-0523 / The Dow Chemical Company / 2023 / https://www.glycolsales.com.au/wp-content/uploads/2025/07/TDS-dowfrost-lc-heat-transfer-fluid.pdf

Shell, “Shell Immersion Cooling Fluid S5 X” technical data sheet, v1.5 / Shell / 2021 / https://www.shell-livedocs.com/data/published/en/615128f8-db18-4709-bb03-10bf668eb6eb.pdf

3M, “3M to Exit PFAS Manufacturing by the End of 2025” / 3M / December 20, 2022 / https://news.3m.com/2022-12-20-3M-to-Exit-PFAS-Manufacturing-by-the-End-of-2025

Schneider Electric (Gregory Alexander, Motivair by Schneider Electric), “What coolant should you use for data center liquid cooling? PG25, dielectric fluids, and ASHRAE recommendations” / Schneider Electric Blog / August 14, 2026 / https://blog.se.com/datacenter/2026/08/14/what-coolant-for-data-center-liquid-cooling-pg25-dielectric-fluids-ashrae-recommendations/

LibreTexts Chemistry, “S2: Properties of Water” (data from CRC Handbook of Chemistry and Physics, 84th Edition, 2004) / LibreTexts / updated November 3, 2020 / https://chem.libretexts.org/Ancillary_Materials/Reference/Reference_Tables/Solvents/S2%3A_Properties_of_Water

Heonsu Nam
Heonsu Nam runs Data Center Technology Lab. His day job is running a small career-education business in South Korea; this site is where he studies data center cooling in public. He is not a mechanical engineer, so every article starts from the primary documents (ASHRAE, Open Compute Project, vendor data sheets) and shows the arithmetic so readers can check it. Spot an error? Please use the Contact page.
함께 보면 좋은 글

댓글 0

첫 댓글을 남겨보세요.