Case Studies in Thermal Engineering 50 (2023) 103472

Contents lists available at ScienceDirect

Case Studies in Thermal Engineering

journal homepage: www.elsevier.com/locate/csite

Closed loop liquid cooling of high-powered CPUs: A case study on

cooling performance and energy optimization
F.M. Naduvilakath-Mohammed a, b, c, *, R. Jenkins b, G. Byrne a, A.J. Robinson a, b, c
a
Department of Mechanical, Manufacturing & Biomedical Engineering, Parsons Building, Trinity College Dublin, Ireland
b
Nexalus Labs, Water St, Loughanalla, Castlepollard, Co Westmeath, Ireland
c
CONNECT, Dunlop Oriel House, Trinity College Dublin, Ireland

A R T I C L E I N F O A B S T R A C T

Keywords: This case study presents an experimental investigation of closed loop liquid cooling of CPUs. The
CPU cooling cooling system implements a pumped liquid supplying a cold plate fixed to the emulated CPU
Liquid cooling heat source and a remote Air-Cooled Heat Exchanger (ACHE). The Hybrid Liquid-Air (HLA)
Hybrid cooling system employs 25% v/v Arteco-Freecor with distilled water solution as the primary coolant, and
Air-cooling air as the secondary coolant. The thermal-hydraulic performance of the system is characterized
Energy optimization
over a range of operating conditions, including CPU heat loads (12–72 W/cm2), water volumetric
flow rates (0.5–8 L/min), and air velocities (1–3 m/s). In order to provide a performance frame of
reference, evaluations were also conducted with a conventional air-based fan-fin CPU cooling
system. The results indicate that the HLA system significantly outperforms the traditional air-
cooled fan-fin system, with much lower CPU temperatures and feasible power levels exceeding
300 W. Additionally, the study highlights the importance of the liquid flow rate and fan speed on
the overall cooling performance. Finally, an optimal combination of liquid and air flow rates is
identified for minimum energy consumption. Overall, these findings support the conclusion that
conventional air-cooling is reaching its theoretical limit and closed loop liquid cooling has the
potential keep pace with the escalating power levels of modern CPUs. However, the additional
energy cost of the cooling system should be carefully managed in order to maximize overall
system Coefficient of Performance.

1. Introduction
The demand for high-performance computing has led CPU manufacturers to develop increasingly powerful processors. These CPUs
continue to consume more power and generate higher heat fluxes, averaging about 70 W/cm2 with Thermal Design Powers (TDPs)
exceeding 150 W [1]. Intel’s i9 7090XE, 56 Xeon Platinum 8470Q (formerly Sapphire Rapids), and 56 Core Xeon Platinum 9282,
among others, exemplify this trend with power outputs of 280 W (overclocked), 350 W, and 400 W respectively [2]. Managing the heat
generated by these CPUs has become a significant and important challenge, since effective cooling solutions are required to prevent
thermal throttling or damage to the CPU.
A simplified thermal resistance network of a CPU packaging is shown in Fig. 1. The core to coolant thermal resistance consists of
five resistances in series, divided into two levels: level one includes the resistance Rdie due to the conduction through the semiconductor
die, the resistance RTIM1 across the Thermal Interface Material (TIM) between die and heat spreader, and the spreading resistance RIHS

* Corresponding author. Department of Mechanical, Manufacturing & Biomedical Engineering, Parsons Building, Trinity College Dublin, Ireland.
E-mail address: [email protected] (F.M. Naduvilakath-Mohammed).

https://doi.org/10.1016/j.csite.2023.103472
Received 15 July 2023; Received in revised form 22 August 2023; Accepted 7 September 2023
Available online 13 September 2023
2214-157X/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

within the Integrated Heat Spreader (IHS); level two consists of resistance RTIM2 across the TIM between IHS and base of the heat
exchanger, and the resistance RHEX associated with the heat exchanging technology to the coolant. Among these, the first-level
resistance is challenging to regulate, as it falls under the jurisdiction of the chip manufacturer. Nevertheless, the resistances at the
second level, in particular RHEX , can be engineered by electronic packagers to reduce RTotal .
The majority of CPUs, whether in PCs or servers, are directly cooled by fan-fin heat sinks. These heat sinks can have duct (parallel)
or impinging (normal) air flow configurations [3]. Desktop PCs typically use impingement cooling, while servers use ducted cooling.
As CPUs push the boundaries of performance and power consumption, the limitations of air cooling become increasingly evident. The
cooling threshold for air cooled fan-fin heat sinks can be as low as 40 W/cm2 [4]. Despite efforts to optimize heat sink designs and
enhance fan performance, there exists a critical point where air cooling can no longer effectively cope with the escalating heat output
of high-performance CPUs. This limitation can push power-hungry CPUs to their thermal limits, putting system performance, stability,
and reliability at risk. The limitation of air cooling becomes more pronounced when considering the non-uniform heat distribution of
modern CPUs [5]. In addition to the heat transfer limitations of fan-fin air cooling, other critical factors like reliability, leakage current,
and harsh operating conditions (especially at elevated temperatures), have emerged as strong motivators for the adoption of liquid
cooling technology. Despite this, adoption rates have been slow. While air cooling remains a practical and cost-effective solution for
many everyday computing needs, it is important to recognize its limitations in the face of ever-evolving CPU thermal challenges. The
pursuit of cutting-edge performance demands high performance thermal management technologies, and liquid cooling is emerging as
an obvious choice for modern high-powered CPUs.
Motivated by the superior thermophysical properties of liquids, single-phase liquid cooling first emerged in the mid-1960s as a
means to cool large mainframe computers. However, it was not until 1981 that Tuckerman and Pease [6] introduced the concept of
direct-to-chip liquid cooling, using microchannels to enable heat flux removal rates exceeding 790 W/cm2. However, with the advent
and uptake of Complementary Metal Oxide-Semiconductor (CMOS) transistors in the 1990s, the computing industry shifted its
attention away from liquid cooling. This was primarily due to the reduced heat generated by CMOS technology, which made air
cooling a feasible and cost-effective solution. Nonetheless, in recent years, there has been a renewed interest in liquid cooling tech­
nologies due to the substantial increase in power densities of transistors and the realization that air cooling is approaching its theo­
retical limit.
Microchannel-based liquid cold plates have been extensively researched as a high potential cooling solutions for high heat flux
electronics, primarily owing to the fact that smaller hydraulic diameters significantly increase the convective heat transfer coefficient,
whilst the interconnected microchannels create high surface area density [7]. With a relatively slow start, there has been renewed
interest in microchannel heat sinks for cooling of electronics since the early years of this century, and research efforts have centred
around improving heat transfer capabilities while managing pressure drop penalties. Various concepts have been explored, including
considerations of channel geometry, flow obstructions, wavy and zig-zag channels, and introducing periodic pinching and channel
roughness [8–11]. This is not an exhaustive inventory of designs conceived to enhance the thermal-hydraulics of microchannel heat
exchangers, yet serves as a testament to the availability of diverse design alternatives for engineering such technologies. Research
highlights include the studies of Colgan et al. [12] who achieved high overall heat transfer coefficients >200 kW/m2 K, and reported
heat flux dissipation >1000 W/cm2 [13]. Nevertheless, the utilization of reduced channel sizes in microchannel heat sinks results in
the trade-off of higher pressure drop and hydraulic power requirements [14]. Also, temperature nonuniformities can arise in
microchannel heat exchangers due to the sensible heating of the fluid along the channel length, nor are they well-suited for managing
hotspots [14]. The presence of temperature gradients in electronics has long been known as detrimental to their reliability due to the
thermal stresses induced by differential thermal expansion [15].
Alternatively, liquid microjet array impingement is a complimentary approach to microchannels as they achieve remarkably high
heat transfer rates without the need for surface area enhancements. For example, in Michna et al. [16] a microjet array dissipated a
heat flux of 1100 W/cm2, albeit from a very small target area. Similar to microchannel flows, the convective heat transfer coefficient
increases as the size of the microjets decreases [17,18]. Despite being relatively less explored in the available literature, microjets are
emerging as a formidable contender for high heat flux cooling applications due to their ability to generate substantially higher
convective heat transfer coefficients compared to traditional channel-based designs, and generally with lower pressure drop penalties

Fig. 1. Thermal resistance network of a CPU Package.

2
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

[19]. The comparatively high convective heat transfer coefficients are attributed to the high velocity of the impinging jets, which
results in a high stagnation heat transfer coefficient at the heated surface. By dispersing coolant across the entire heated surface, jet
arrays also have the potential to greatly reduce undesired surface temperature gradients. Furthermore, microjet arrays can be designed
to provide targeting of hotspots to improve surface temperature uniformity [19]. However, the heat transfer between neighbouring jets
in the array tends to degrade rapidly, presenting an engineering challenge in achieving high cooling effectiveness over large surface
areas at reasonable volumetric flow rates [20].
Hybrid microjet-microchannel systems have also been studied. Here, the liquid is discharged through the perforated orifice plate as
multiple microjets into a microchannel feature on the cold plate, thus capitalising on the merits of each technology to achieve the high
cooling intensity whilst mitigating high temperature gradients and hydraulic penalties. For example, Robinson et al. [21], achieved
effective conductance levels >400 kW/m2 K, at a moderate pressure drop (<100 kPa).
In a closed loop configuration, ensuring optimal performance relies not only on the cold plate heat sink but also on the integration
of a reliable and compact pump to facilitate fluid circulation within the loop. Additionally, a liquid-air heat exchanger is required to
dissipate the captured heat from the cold plate. Furthermore, improved heat transfer provided by the liquid cold plate comes at the
price of the pressure drop and associated pumping power. This, and any additional fan power requirements, must be considered in the
context of the overall energy consumption of the cooling system.
From a system integration perspective, existing research has primarily focused on the CPU alone, whereas the full source-to-sink
thermal-hydraulics has been largely overlooked. Some studies aimed to address this gap by investigating the overall thermal resistance
of liquid cooling systems, including the remote heat exchanger and the interplay of all system components. A compact CPU cooling
system was studied by Bintoro et al. [22], comprising a single water jet impingement device and mini-channel remote heat exchanger,
achieving an overall source-to-sink thermal resistance of approximately 0.35 K/W. Similarly, Liu et al. [23] and Chang et al. [24] used
microjets and microchannel heat sinks respectively, coupled with a remote heat exchanger to achieve overall thermal resistances of
about 0.17–0.23 K/W. Efforts by Whelan et al. [25] to design and develop a full closed loop CPU liquid cooling system achieved an
overall thermal resistance of 0.18 K/W, with a low pumping power of 1.5 W.
Considering the above discussion, the current study is motivated by the recognition that high-powered CPUs will necessitate
cooling solutions that surpass the inevitable limitations of conventional fan-fin air cooling. Single-phase liquid cooling emerges as a
very high potential option from technical and practical standpoints. However, existing literature highlights knowledge gaps con­
cerning the system-level assessment of closed loop liquid-based cooling technology. These gaps encompass full system-level thermal-
hydraulic performance characteristics across a range of operating conditions, as well as energy management considerations, partic­
ularly in terms of characterizing and optimizing the energy penalty associated with the full cooling system. Thus, the objective of this
research is to conduct a comprehensive case study on CPU closed loop liquid cooling, encompassing the thermal-hydraulics from the
heat source to the heat sink. By performing a thorough parametric study across a sufficient range of system parameters, the research
aims to fill gaps in the existing literature by gaining a deeper understanding of the parameter sensitivity, and how this influences
overall performance of the end-to-end systems. Since this investigation compares the performance characteristics with traditional fan-
fin technology, it also presents electronics packagers with missing information, as well as a methodology and characterisation pro­
cedure, for making practical decisions when evaluating the limitations of conventional air cooling. Finally, the investigation extends to
optimizing the energy cost associated with closed loop liquid cooling.

2. Experimental apparatus and data reduction

2.1. Thermal test vehicle with liquid-cooled cold plate and remote air-cooled heat exchanger
One key aim of this investigation is to evaluate the thermal-hydraulic performance characteristics of a pumped Hybrid Liquid-Air
(HLA) system consisting of a cold plate and remote Air-Cooled Heat Exchanger (ACHE). To achieve this, an instrumented flow loop
facility that emulates the practical implementation of this type of cooling solution was constructed. As Fig. 1 indicates, the flow loop
has three region of interest – the Thermal Test Vehicle (TTV), the primary (liquid) coolant loop, and the ACHE. As will be discussed, the
TTV is comprised of a variable heat source which is cooled by a liquid cold plate. The present study employs a coolant mixture
comprising 25% v/v Arteco-Freecor FTC® and 75% v/v distilled water as the primary coolant. This coolant mixture was selected to
provide freeze protection and inhibit corrosion and microbial growth compared to water [26].
The primary loop includes a SAER CF gear pump with an AC10 series Parker Hannifin single phase inverter drive that controls the
pump speed to the desired liquid flow rate. The flow rate is measured by a non-invasive UF25B Cynergy ultrasonic flow meter with
uncertainty of ± 3 % of reading. Pressure drop measurements are taken across both the cold plate and the ACHE with a 26PCDFA6D
Honeywell differential pressure transducer with uncertainty of ±0.25%. Similarly, temperature differentials are taken across both the
cold plate and the ACHE with T-Type thermocouples, which were inserted into the flow via compression fittings. The temperature
probes were calibrated against a platinum RTD reference probe to an uncertainty of a ±0.2 ◦ C.
A Thermal Test Vehicle (TTV) was constructed to represent a CPU heat source. As depicted in Fig. 3(a), the TTV was machined from
single block of ultra-pure OFE copper, with a lower neck region of Adie = 20 × 20 mm2 square cross section, roughly the area of higher
powered (>150 W) commercial CPU dies (e.g. Intel i9 7980XE), which tend to be marginally larger than their lower-powered
counterparts. Three OMEGALUX® CS series stainless steel sheath cartridge heaters of 6.35 mm diameter, that could supply a
maximum total power of 100 W each, were inserted into the bottom of the 60 mm long lower neck section to provide heat to the upper
40 mm × 40 mm cross section and 2.5 mm thick overhang section, which represents a CPU Integrated Heat Spreader (IHS). The TTV
section is then housed in a 3D printed case lined with insulation to minimized heat losses and provide a mounting fixture for the
attached liquid-cooled cold plate.

3
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

The measurement of the heat flux from the lower neck to the IHS, as well as the temperature of the tip of the neck region, here
representing the CPU die temperature, is facilitated by three 1.0 mm diameter T-Type thermocouples spaced at equal intervals along its
length and inserted into drilled holes to a depth of 10 mm. As Fig. 3(b) illustrates, this allows the determination of the temperature
gradient at the base of the IHS, dT/dx, such that the associated heat transfer to the IHS section can be determined from,
dT
Q = − kCu Adie (1)
dx

The temperature of the neck-IHS interface, T4 (hereinafter Tdie ), represents the temperature of the CPU die and is determined from
linear extrapolation,
( )( )
Q Δx3
Tdie = T3 − (2)
Adie kCu
The IHS above the heat source was cooled by a commercial cold plate (Nexalus Enflux™). As Fig. 2 indicates, this cold plate consists
of a plastic manifold that forces the liquid through an array of 0.7 mm orifices, creating an array of high velocity microjets that impinge
on a nickel-plated copper baseplate. The cold plate was mounted to the TTV via four bolts tightened with a torque wrench to ensure
correct and even pressure. A layer of Nexalus Hydronex™ thermal grease is placed between the IHS and the cold plate to reduce contact
resistance, typically referred to as TIM2.
With the heat transfer to the IHS measured, the overall die-to-coolant thermal resistance can be estimated as,
ΔTcp
Rd− c = (3)
Q

As suggested by Webb [27], the temperature difference across the cold plate, ΔTcp , is calculated as the Log-Mean Temperature Dif­
ference (LMTD),
Tout − Tin
ΔTcp = ( ) (4)
ln TTdiedie−− TTout
in

It must be noted that the overall thermal resistance includes the resistance contributions of the IHS, TIM2 and the cold plate heat
exchanger. In a commercial CPU package, additional thermal resistances associated with TIM1 needs to be included for accurate
estimation of the die temperature.
Consistent with modern liquid cooling technology currently deployed in high-performance desktop/workstation computing sys­
tems, the heat rejection to the air was facilitated via a remote air-liquid heat exchanger. In this investigation, the Nexalus NXQ 360 fan-
cooled liquid-air flat tube heat exchanger of outer dimensions 400 mm × 130 mm × 60 mm, fitted with three 120 mm high static
pressure fans, was used to transfer the heat from the liquid to the air. The heat exchanger is equipped with 16 flat copper tubes, each 2
mm wide, spanning its entire length. The copper fins of the radiator are non-louvered and have a density of 18 fins per inch (FPI).
Thermocouples were positioned in the entrance and exit air streams in order to evaluate the temperature rise across the air-side of the
heat exchanger. These measurements, along with those of the water-side, allow for the calculation of the overall thermal resistance of
the heat exchanger,

Fig. 2. Wireframe diagram of the liquid cooling system using air cooled heat exchanger system.

4
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

Fig. 3. (a) Schematic of the Thermal Test Vehicle (TTV) with mounted liquid-cooled cold plate, and (b) illustration of heated meter bar and the resulting linear
Temperature vs. Distance plot.

1 ΔThex
Rhex = = (5)
(UA)hex Q

where ΔThex is the LMTD of the crossflow liquid-air heat exchanger and is calculated by,
( ) ( )
Thex,out − Tair,in − Thex,in − Tair,out
ΔTACHE = ( ) (6)
(Thex,out − Tair,in )
ln T − T
( hex,in air,out )

The fan speed was controlled by varying the supply voltage with a desktop DC power supply over the range of 6–14 V. The penalty
associated with the power required for the liquid-side of the cooling system requires determination of the hydraulic pumping power
across the cold plate and ACHE,
( )
Ẇ pumping = V̇ ΔPcp + ΔPACHE (7)

where V̇ is the liquid volumetric flow rate and ΔP is the associated pressure drop. It is important to note that the pressure drops arising
from the valves, T-joints, and flow meter are not considered here as these would be absent in a real-life application. The total pumping
power along with the total fan power,

Ẇ fans = IV (8)

allows for the estimation of the overall system Coefficient of Performance (COP),
Q
COPACHE = (9)
Ẇ pumping + Ẇ fans

Fig. 4. Wireframe diagram of the impingement flow Fan-Fin Air Cooled (FFAC) system.

5
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

2.2. Fan-fin CPU cooler

In order to establish a performance frame of reference for comparison with the liquid-cooled system described above, a commercial
off-the-shelf Fan-Fin Air Cooled (FFAC) system was also characterized. The FFAC used was typical of most commercial impingement
type heat sinks, with aluminium fins measuring 95 mm × 98 mm × 70 mm and equipped with a 92 mm fan. The heat sink was affixed to
the Integrated Heat Spreader (IHS) of the TTV via mechanical fastening, with the same layer of thermal grease (TIM2) placed between
the IHS and the aluminium heatsink base, as depicted in Fig. 4. The heat sink had a total of 35 fins, each of 0.6 mm thickness and 34.5
mm height, with 1.8 mm spacing between them.
In characterizing the thermal performance of the FFAC, the source-to-sink thermal resistance (RFFAC ) is calculated as,
ΔTFFAC
RFFAC = (10)
Q

where ΔTFFAC is the LMTD of the cold plate heatsink and is calculated by
Tair,out − Tair,in
ΔTcp = ( ) (11)
T − Tair,in
ln Tdiedie− Tair,out

By utilizing a benchtop DC power supply, the fan speed was regulated, and the total power was computed similar to Eq. (8), such
that the system’s Coefficient of Performance (COP) can be determined from,
Q
COPFFAC = (12)
Ẇ fans

2.3. Uncertainty analysis

The uncertainties associated with the primary measurement instruments are provided in Table 1. To determine the propagated
uncertainty of calculated quantities, standard methodologies outlined by Taylor [28] were followed, assuming that the uncertainties
were random and independent. In cases where a calculated quantity Z relied on measured quantities X, Y, and possibly others, the
uncertainty δZ was determined using the following prescribed procedures.
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2 ( )2
δX δY
δZ = |Z| + +… (13)
|X| |Y|

The uncertainty associated with the rate of heat transfer through the heated meter bar system requires some special attention, as it
depends not only multiple temperature and spatial errors, but also the regression fitting to these to determine dT/dx. Similarly, this
error propagates to the estimation of Tdie , which must also be quantified. To account for this, the Monte-Carlo calculation method first
described in Kempers et al. [29] and later detailed in Kempers and Robinson [30] was implemented.

3. Results and discussion

3.1. Thermal characterisation
3.1.1. Fan-fin air cooling system
Fig. 5 shows that, for the FFAC system, as the heat flux supplied from the TTV increases from 12.25 W/cm2 to 72 W/cm2 the net
thermal resistance remains relatively constant within their absolute uncertainties. Increasing the air velocity across the heat sink
lowers its thermal resistance from about 0.32 K/W to 0.27 K/W, indicating marginally improved heat transfer. Fig. 6 shows the ex­
pected cooling performance of the FFAC system under various heat loads for an ambient temperature of 35 ◦ C, the standard room
temperature recommended by Intel [31]. Under extreme loads, the average die temperatures can reach as high as 80–85 ◦ C, and this is
regarded as a ‘safe’ operating temperature limit [32]. The throttling temperature limit may differ depending on the CPU type, typically
falling within the range of 100–105 ◦ C [33]. In this study, the safe and throttling temperature limits of 83 ◦ C and 102 ◦ C respectively,
are chosen when comparing cooling technologies. To predict the die temperature closer to what would be expected in a CPU package,
an approximation for the TIM1 thermal resistance, using the RTIM1 = 0.1 K–cm2/W, was included [4].
For a normal fan speed of 1.8 m/s, the safe die temperature limit of 83 ◦ C was breached at approximately 35 W/cm2, which is close
to that reported in the literature [4]. The temperature at the maximum heat flux studied (72 W/cm2) was at least 53 K higher than the
safe limit, and even well above the throttling temperature of 102 ◦ C [33]. Consistent with the thermal resistance behaviour, there is a

Table 1
Uncertainties of measured quantities.

Measurement Instrument Absolute Uncertainty

Temperature T-type Thermocouple ±0.2 ◦ C

Flowrate Ultrasonic Flowmeter ±3%
Distance Vernier Calliper ±0.0005 m
Pressure Drop Pressure Sensors ±0.25%
Air Velocity Vane Anemometer ±2%

6
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

Fig. 5. Thermal resistance of fan-fin based air cooling system with varied heat loads; Tambient = 20◦ C.

Fig. 6. Expected cooling performance of a standard Fan & Fin Air Cooled system at maximum fan speed with varying heat loads for Tambient = 35 ◦ C and RTIM1 = 0.1 K-
cm2/W.

noticeable yet marginal effect of air velocity on the die temperature, which diminishes with increased velocity. Considering the
diminishing gain in the heat transfer, together with the escalation in fan noise and power consumption, it is not worthwhile to increase
the air speed beyond approximately 1.8 m/s.
As Fig. 7 indicates, increasing the heat flux results in an increase in the Coefficient of Performance (COP) of the system, ranging
approximately from ~15 to 90 for settings that achieve lower than the throttling temperature. This result is expected since, despite the
increase in heat transfer rate, the fan power required to cool CPU was kept constant. However, as Fig. 7 shows, higher fan speeds lead to
a significant decrease in the COP. This occurs because, even though the heat flux remains constant, the fan power consumption nearly
doubles when the air velocity changes from 1.3 m/s to 2.1 m/s, resulting in the COP being halved. Upon closer examination of the
temperature and COP plots, it was observed that surpassing an air velocity of 1.8 m/s does not provide substantial cooling benefits.
Instead, it leads to increased power consumption, lower COP values, and heightens fan noise. This highlights a key point which has not
been addressed sufficiently in the literature: a balance needs to be struck between cooling requirements and the associated penalties,
like energy consumption and noise. Each aspect must be considered holistically in the development of a cooling system.

3.1.2. Hybrid Liquid-Air (HLA) cooling system

Fig. 8 shows that for the selected nominal settings (max fan speed & V̇ = 4 L/min), the thermal resistance of the cold plate and
ACHE remains almost constant, as the heat flux increases from 12.25 W/cm2 to 72 W/cm2. It is worth noting that the thermal resistance

7
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

Fig. 7. Power Consumption & Coefficient of Performance (COP) profile of the fan-fin heat sink; Tambient = 20 ◦ C.

of the cold plate is roughly ten times that of the ACHE, which supports the focus on cold plate research in the literature. Notably, the net
thermal resistance of the liquid cooling system is about 0.11. K/W, which is three times lower than the FFAC, which is significant.
The expected die temperature with increasing heat flux are plotted in Fig. 9, again considering Tambient = 35 ◦ C and approximate
TIM1 thermal resistance. As the primary and secondary coolant temperatures experience only slight increases over the range of
considered heat fluxes, Tdie increases linearly with the heat transfer rate, attributable to the constant source-to-sink thermal resistance.
The HLA system maintains temperatures well within the safe operating limits for the full range of high heat fluxes tested, and throttling
is not expected until power levels of close to 450 W. When comparing the performance of the HLA and FFAC systems, it was found that
the HLA system is capable of cooling the die by at least 60 K more effectively than the fan-based system for the highest tested heat flux
of 72 W/cm2.
Fig. 10 plots the HLA system pressure drops at varying heat load for the nominal flow rate of 4 L/min across both the cold plate and
ACHE. The cold plate demonstrates an expectedly higher pressure drop compared to ACHE (approximately 16 kPa compared to 2 kPa),
though these combine to a pumping power of only 1.1 W. It is worth noting that the fan power at the ACHE for this maximum air
velocity is nearly six times greater than the hydraulic power necessary to circulate the liquid in the loop, as shown in Fig. 11. Therefore,
previous research efforts that focused on enhancing cold plate heat transfer capabilities whilst managing the pressure drop are
justified.
As seen in Fig. 11, the COP of the system is observed to rise linearly, as expected, from approximately 5 to 36 over the range of TTV
powers tested. A slight decrease in hydraulic power is observed due to the lower pressure drop in the liquid loop caused by a slightly
lower coolant viscosity at the higher operating temperatures, as seen in Fig. 10. Compared to the FFAC, the COP of the HLA system is
approximately 2.5× lower for settings that achieve lower than the throttling temperature. However, this penalty must be considered in
the context of the benefits associated with the lower operating CPU temperatures and extended power range afforded by the HLA

Fig. 8. Thermal Resistance of HLA cooling system with varied heat loads: Tambient = 20 ◦ C.

8
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

Fig. 9. Expected cooling performance of HLA cooling system at varying heat loads for a nominal flow rate of 4 L/min and fan speed of 3 m/s at Tambient = 35 ◦ C and
RTIM1 = 0.1 K-cm2/W.

Fig. 10. Pressure drop curves of the HLA system for varying heat fluxes at constant coolant flow rates: Tambient = 20 ◦ C.

system.

3.2. Liquid flow characterisation – HLA system

Fig. 12 shows the variation of thermal resistance with liquid flow rate for a fixed maximum heat flux and fan speed. The plot shows
that as the liquid flow rate increased from 0.5 L/min to 8 L/min, the thermal resistance decreased, though its effect is much more
pronounced for the cold plate compared to the ACHE. However, beyond around 4 L/min, the flow rate does not have significant effect
on the thermal resistance. This is also noticeable from the temperature plots in Fig. 13, whereby any increase in coolant flow rate
beyond 4 L/min has diminishing influence on the die temperature, which are all predicted to be below the safe operating temperature
for this maximum heat flux rest case. This diminishing return with liquid flow rate is expected since (i) jet array impingement heat
transfer increases as Nu∝Rend , with n ~ 0.5 [17], and (ii) as the convective resistance decreases, those of the IHS and TIM2 become
increasingly dominant.
However, as shown in Fig. 14, the pressure drop across the cold plate and ACHE increases with the expected parabolic profile, with
the former dominating the total system pressure drop and thus hydraulic power. As such, the diminishing gains in heat transfer
associated with flow rates beyond about 4 L/min has associated with it increasingly severe hydraulic penalties. Again, this illustrates
that, when considered as a full system, careful design consideration must be made to ensure adequate performance with managed
hydraulic penalties. This is particularly important considering the impact on the overall energy dynamics of the HLA system. This is
displayed in Fig. 14, where increasing flow rates from 4 L/min to 8 L/min leads to about a 3.5-fold increase in the hydraulic power,

9
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

Fig. 11. Power Consumption & Coefficient of Performance (COP) profile of the HLA system; Tambient = 20 ◦ C.

Fig. 12. Thermal Resistance of HLA system for varying liquid flow rates measured at constant heat flux and air velocity for Tambient = 20 ◦ C.

which negatively affects the system COP, as shown in Fig. 15, with little improvement in the cooling performance (Fig. 12). However,
this is moderated to some degree by the fan power (~6.6 W), which is notably higher than the hydraulic power for liquid flow rates up
to 4 L/min. This causes the COP to level-off at lower liquid flow rates, and to a lower value for decreasing heat input levels (see Fig. 15).

3.3. Fluid characterisation – air side

Keeping the heat flux constant (72 W/cm2) and liquid flow rate at its nominal 4 L/min, the air velocity (fan speed) was varied
between 1 and 3 m/s to investigate its overall influence on system performance. As evident from Fig. 16, increasing the air velocity has
much less significant influence on the ACHE thermal resistance compared with the effect the liquid flow rate has on the cold plate
resistance (Fig. 12). It is thus unsurprising that the predicted die temperature is nearly constant with increasing fan speed, as seen in
Fig. 17. The same cannot be said about the power consumption and related COP, which depicted in Fig. 18, shows the system COP
decreasing steeply with increased fan speed. This is due to the increased power consumption of the fan, which increases about 5.5-fold.
Notably, Fig. 18 shows that the fan power can be almost as low as the pumping power at the lowest fan speed setting, and this occurs
with only a minor drop in the die temperature (Fig. 17).
This investigation of closed loop liquid cooling on a high-powered CPUs showed significantly higher cooling capacity than the
traditional FFAC system. For a given setting of the fans and pump, the overall thermal resistance is relatively insensitive to the applied
heat flux. Increasing the liquid flow rate improved the heat transfer to a point, after which there was diminishing returns in terms of
higher power consumption. Lastly, increasing the fan speed had minimal impact on the overall heat transfer while significantly

10
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

Fig. 13. Expected Cooling Performance curves of HLA system for varying liquid flow rates at constant heat flux and air velocity at Tambient = 35 ◦ C and RTIM1 = 0.1 K-
cm2/W.

Fig. 14. Pressure drop plots of HLA for varying liquid flow rates at constant heat load; Tambient = 20 ◦ C.

increasing power consumption and decreasing the COP. These findings highlight the importance of effective management of the fluid
movers to optimize energy dynamics in liquid cooling systems. Optimizing power consumption of the HLA system is crucial, as it
remains the main drawback compared to conventional FFAC systems, highlighting the need for a comprehensive power optimization
study.

3.4. Cooling energy optimization

To illustrate the potential for optimizing the cooling energy requirement of the HLA system, cases are considered where the fluid
movers are controlled in such a way to maintain a constant die temperature. Practically, CPUs operate from idle to their TDP, and
maintaining a constant die temperature would be beneficial since it would reduce thermal cycling over large computational load
swings. Two case studies are considered, one for maintaining Tdie = 75 ◦ C and a more aggressive case where Tdie = 55 ◦ C, each for the
more common situation where Tambient = 20 ◦ C. The full heat flux range from 6 W/cm2 to 72 W/cm2 is then considered for the primary
liquid coolant flow rates between 0.5 L/min and 8 L/min and the secondary air coolant velocities of 1–3 m/s.
Fig. 19 (a) and (b), plot the total cooling power consumption to maintain die temperatures of 75 ◦ C and 55 ◦ C respectively, over the
range of die heat fluxes and as a function of the pump and total fan flow required to sustain the setpoint temperature. To effectively
maintain the constant die temperature at a given heat flux, Fig. 19 shows that the minimum cooling power consumption adapts to a
higher liquid flow rate and lower air velocity (fan speed). For example, considering the 75 ◦ C/60 W/cm2 case, the same overall cooling
performance can be achieved with a liquid flow rate of 0.52 L/min and air velocity of ~3.0 m/s, as it can with a higher liquid flow rate
of 2.2 L/min and lower air velocity of 1.0 m/s. Here the total cooling power drops from 5.2 W to 1.8 W, which is better than what can

11
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

Fig. 15. COP plots of HLA system for three different heat fluxes with varying liquid flow rates measured at constant air velocity: Tambient = 20 ◦ C.

Fig. 16. Thermal Resistance of HLA system for varying air velocities measured at constant heat flux and liquid flow rate for Tambient = 20 ◦ C.

be achieved by the FFAC system, despite the fact that the FFAC system cannot cool sufficiently at this heat flux level (Fig. 6). Referring
to the COP surface plot in Fig. 20(a), the resulting gain for this case is from 52.1 to 156.4, which is significant. For the more aggressive
cooling case of 55 ◦ C/60 W/cm2, Fig. 19(b) and Fig. 20(b) show that it is optimal to operate the HLA system with the liquid and air fluid
movers set at 4 L/min and 1 m/s respectively, resulting in an energy consumption of 3.3 W (COP ~ 75). In contrast, utilizing the
maximum air speed of 3 m/s, a liquid flow rate of only 1.9 L/min is required, though the energy consumption is 2 times higher, with a
commensurate drop in COP. Therefore, opting for lower fan speed requires higher liquid flow rates to deliver equivalent cooling
performance, though consuming comparatively less overall power, which improves the COP. It is interesting to note that the maximum
air-speed of 3 m/s is never required for the two setpoint test cases, and apart from the highest heat flux, only ~1 m/s (lowest fan speed
setting) is generally required for minimum cooling power and maximum COP. Thus, the temperature can be regulated, at minimum
cooling power expense, by controlling the liquid flow rate. In doing so, high COPs can be achieved for the range of CPU heat fluxes
between 24 W/cm2 and 72 W/cm2, ranging between COP ~ 97 and 150 for the Tdie = 75 ◦ C setpoint, and COP ~ 60–73 for the Tdie =
55 ◦ C setpoint. Importantly, this is comparable to a FFAC cooling system, whilst significantly expanding the maximum TDP capability
of CPUs.

4. Conclusions
This case study aimed to investigate the cooling performance of a closed loop liquid cooled system for advanced thermal

12
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

Fig. 17. Temperature profile of the HLA system for varying air velocities measured at constant heat flux and liquid flow rate at Tambient = 35 ◦ C and RTIM1 = 0.1 K-
cm2/W.

Fig. 18. Power consumption of the liquid cooling system for varying air velocities measured at constant heat flux and liquid flow rate of 4 L/min: Tambient = 20 ◦ C.

Fig. 19. HLA power consumption surface plot for maintaining the die temperature at (a) Tdie = 75 ◦ C and (b) Tdie = 55 ◦ C, by varying liquid flow rate & air velocity
with Tambient = 20 ◦ C and RTIM1 = 0.1 K-cm2/W.

13
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

Fig. 20. HLA COP surface plot for maintaining the die temperature at (a) Tdie = 75 ◦ C and (b) Tdie = 55 ◦ C, by varying liquid flow rate & air velocity with Tambient =
20 ◦ C and RTIM1 = 0.1 K-cm2/W.

management of high-performance CPUs. For an emulated 2 × 2 cm2 CPU die and range of input power and heat fluxes up to 288 W and
72 W/cm2 respectively, the experimental investigation measured key cooling performance metrics of a Hybrid Liquid-Air (HLA)
system, such as die temperature and thermal resistance, over ranges of liquid flow rates (0.5 L/min – 8 L/min) and fan speeds (air
velocities of 1 m/s to 3 m/s). The cooling pressure drop, and power penalty was also measured. To provide a comparison, a traditional
Fan-Fin AirCooled (FFAC) system was also tested, and the following conclusions can be drawn:
(1) The HLA system outperforms the FFAC system in terms of source-to-sink thermal resistance and associated lower CPU
temperatures.
(2) The feasibility limit of the FFAC system tested was around 42 W/cm2 (168 W), whilst the HLA system has an expected capacity
to cool over 112.5 W/cm2 (450 W), illustrating that liquid-based cooling can support the escalating CPU power levels, whilst
traditional air cooling cannot.
(3) The power requirement of the two fluid movers (pump and fans) in the HLA system can greatly exceed the FFAC system, if not
correctly managed.
(4) With two fluid movers, maintaining a die temperature can be achieved by controlling either or both the pump and fan speeds,
and the power cost penalty and overall COP of the system are sensitive to this choice.
(5) By controlling the pump flow rate and fan speed, a constant setpoint die temperature can be maintained over a range of CPU
heat fluxes. Importantly, this can be achieved in such a way as to minimize the cooling power and thus maximize the system
COP.
(6) For a given CPU heat flux, lowering the fan speed and increasing the liquid flow rate to achieve a die temperature setpoint
minimizes cooling power and maximizes COP.
(7) For the most aggressive case investigated, with a CPU heat flux of 72 W/cm2 (288 W) and CPU die temperature of Tdie = 55 ◦ C,
the optimum power for the range of fluid flow rates tested is only 5.4 W, with an impressive COP = 56, illustrating that industry
concerns about the additional energy cost of liquid cooling can be largely addressed by engineering suitable control systems to
manage the fluid movers.
The demonstrated superiority of the HLA system over the conventional FFAC system for the tested CPU heat transfer rates un­
derscores the importance of advanced liquid cooling solutions for energy efficient and reliable operation of modern CPUs and other
processors, such as GPUs. Future research will investigate means to further increase the cooling capacity of liquid-based cooling of
high-powered processors by addressing the liquid coolant temperature in an HLA system. Being limited by the ambient air temper­
ature, including a miniature vapour-compression refrigeration system has the potential to provide sub-ambient liquid coolant tem­
peratures, thus increasing the cooling capacity of liquid-based thermal management systems.

Authors statement
Fazeel Mohammed N. M.: Conceptualization, Methodology, Data Curation & Investigation, Richard Jenkins: Product and process
realization, Gerard Byrne: Technical Knowledge and support, Anthony James Robinson: Supervision, review and editing.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to
influence the work reported in this paper.

Data availability

Data will be made available on request.

14
F.M. Naduvilakath-Mohammed et al. Case Studies in Thermal Engineering 50 (2023) 103472

Acknowledgments
This work was developed with financial support of the IRC-EPS Award EPSPG/2022/347 and CONNECT research centre via Science
Foundation Ireland (SFI) grant number 13/RC/2077_P2.

References
[1] J.W. Elliott, M.T. Lebon, A.J. Robinson, Optimising integrated heat spreaders with distributed heat transfer coefficients: a case study for CPU cooling, Case Stud.
Therm. Eng. 38 (2022), 102354, https://doi.org/10.1016/j.csite.2022.102354.
[2] Intel product specifications, (n.d.). https://www.intel.com/content/www/us/en/ark.html (accessed May 16, 2023).
[3] M. Saini, R.L. Webb, Validation of models for air cooled plane fin heat sinks used in computer cooling, in: ITherm 2002. Eighth Intersociety Conference on
Thermal and Thermomechanical Phenomena in Electronic Systems (Cat. No.02CH37258), 2002, pp. 243–250, https://doi.org/10.1109/
ITHERM.2002.1012464.
[4] M. Saini, R.L. Webb, Heat rejection limits of air cooled plane fin heat sinks for computer cooling, IEEE Trans. Compon. Packag. Technol. 26 (2003) 71–79,
https://doi.org/10.1109/TCAPT.2003.811465.
[5] A.L. Moore, L. Shi, Emerging challenges and materials for thermal management of electronics, Mater. Today 17 (2014) 163–174, https://doi.org/10.1016/j.
mattod.2014.04.003.
[6] D.B. Tuckerman, R.F.W. Pease, High-performance heat sinking for VLSI, IEEE Electron. Device Lett. 2 (1981) 126–129, https://doi.org/10.1109/
EDL.1981.25367.
[7] A. Mohammed Adham, N. Mohd-Ghazali, R. Ahmad, Thermal and hydrodynamic analysis of microchannel heat sinks: a review, Renew. Sustain. Energy Rev. 21
(2013) 614–622, https://doi.org/10.1016/j.rser.2013.01.022.
[8] A.A. Alfaryjat, H.A. Mohammed, N.M. Adam, M.K.A. Ariffin, M.I. Najafabadi, Influence of geometrical parameters of hexagonal, circular, and rhombus
microchannel heat sinks on the thermohydraulic characteristics, Int. Commun. Heat Mass Transfer 52 (2014) 121–131, https://doi.org/10.1016/j.
icheatmasstransfer.2014.01.015.
[9] Y.J. Lee, P.K. Singh, P.S. Lee, Fluid flow and heat transfer investigations on enhanced microchannel heat sink using oblique fins with parametric study, Int. J.
Heat Mass Transfer 81 (2015) 325–336, https://doi.org/10.1016/j.ijheatmasstransfer.2014.10.018.
[10] G. Xie, J. Liu, Y. Liu, B. Sunden, W. Zhang, Comparative study of thermal performance of longitudinal and transversal-wavy microchannel heat sinks for
electronic cooling, J. Electron. Packag. 135 (2013), https://doi.org/10.1115/1.4023530.
[11] Y.L. Zhai, G.D. Xia, X.F. Liu, Y.F. Li, Heat transfer in the microchannels with fan-shaped reentrant cavities and different ribs based on field synergy principle and
entropy generation analysis, Int. J. Heat Mass Transfer 68 (2014) 224–233, https://doi.org/10.1016/j.ijheatmasstransfer.2013.08.086.
[12] E.G. Colgan, B. Furman, A. Gaynes, W. Graham, N. LaBianca, J.H. Magerlein, R.J. Polastre, M.B. Rothwell, R.J. Bezama, R. Choudhary, K. Marston, H. Toy,
J. Wakil, J. Zitz, A practical implementation of silicon microchannel coolers for high power chips, in: Semiconductor Thermal Measurement and Management
IEEE Twenty First Annual IEEE Symposium, 2005, 2005, pp. 1–7, https://doi.org/10.1109/STHERM.2005.1412151.
[13] J. Zhou, X. Cao, N. Zhang, Y. Yuan, X. Zhao, D. Hardy, Micro-Channel heat sink: a review, J. Therm. Sci. 29 (2020) 1431–1462, https://doi.org/10.1007/
s11630-020-1334-y.
[14] W. Qu, I. Mudawar, Experimental and numerical study of pressure drop and heat transfer in a single-phase micro-channel heat sink, Int. J. Heat Mass Transfer 45
(2002) 2549–2565, https://doi.org/10.1016/S0017-9310(01)00337-4.
[15] A.M. Kiper, Impinging water jet cooling of VLSI circuits, Int. Commun. Heat Mass Transfer 11 (1984) 517–526, https://doi.org/10.1016/0735-1933(84)90003-
4.
[16] G.J. Michna, E.A. Browne, Y. Peles, M.K. Jensen, The effect of area ratio on microjet array heat transfer, Int. J. Heat Mass Transfer 54 (2011) 1782–1790,
https://doi.org/10.1016/j.ijheatmasstransfer.2010.12.038.
[17] A.J. Robinson, E. Schnitzler, An experimental investigation of free and submerged miniature liquid jet array impingement heat transfer, Exp. Therm. Fluid Sci.
32 (2007) 1–13, https://doi.org/10.1016/j.expthermflusci.2006.12.006.
[18] A. Robinson, A thermal–hydraulic comparison of liquid microchannel and impinging liquid jet array heat sinks for high-power electronics cooling, components
and packaging technologies, IEEE Trans. On 32 (2009) 347–357, https://doi.org/10.1109/TCAPT.2008.2010408.
[19] D. Nikolic, M. Hutchison, P.T. Sapin, A.J. Robinson, Hot spot targeting with a liquid impinging jet array waterblock, in: 2009 15th International Workshop on
Thermal Investigations of ICs and Systems, 2009, pp. 168–173.
[20] A. Bhunia, C.L. Chen, On the scalability of liquid microjet array impingement cooling for large area systems, J. Heat Transfer 133 (2011), https://doi.org/
10.1115/1.4003532.
[21] A.J. Robinson, R. Kempers, J. Colenbrander, N. Bushnell, R. Chen, A single phase hybrid micro heat sink using impinging micro-jet arrays and microchannels,
Appl. Therm. Eng. 136 (2018) 408–418, https://doi.org/10.1016/j.applthermaleng.2018.02.058.
[22] J.S. Bintoro, A. Akbarzadeh, M. Mochizuki, A closed-loop electronics cooling by implementing single phase impinging jet and mini channels heat exchanger,
Appl. Therm. Eng. 25 (2005) 2740–2753, https://doi.org/10.1016/j.applthermaleng.2005.01.018.
[23] S. Liu, J. Yang, Z. Gan, X. Luo, Structural optimization of a microjet based cooling system for high power LEDs, Int. J. Therm. Sci. 47 (2008) 1086–1095, https://
doi.org/10.1016/j.ijthermalsci.2007.09.005.
[24] J.-Y. Chang, H.S. Park, J.I. Jo, S. Julia, A system design of liquid cooling computer based on the micro cooling technology, in: Thermal and Thermomechanical
Proceedings 10th Intersociety Conference on Phenomena in Electronics Systems, 2006, ITHERM 2006, 2006, pp. 157–160, https://doi.org/10.1109/
ITHERM.2006.1645337.
[25] B.P. Whelan, R. Kempers, A.J. Robinson, A liquid-based system for CPU cooling implementing a jet array impingement waterblock and a tube array remote heat
exchanger, Appl. Therm. Eng. 39 (2012) 86–94, https://doi.org/10.1016/j.applthermaleng.2012.01.013.
[26] Freecor FTC | Arteco Coolants, (n.d.). https://www.arteco-coolants.com/en/products/engine-coolants/x-oat-coolants/freecor-ftc (accessed March 31, 2023).
[27] R.L. Webb, Heat exchanger design methodology for electronic heat sinks, J. Heat Transfer 129 (2006) 899–901, https://doi.org/10.1115/1.2717249.
[28] J.R. Taylor, An Introduction to Error Analysis, University Science Books, Sausalito, Calif., 1997.
[29] R. Kempers, P. Kolodner, A. Lyons, A.J. Robinson, A high-precision apparatus for the characterization of thermal interface materials, Rev. Sci. Instrum. 80
(2009), 095111, https://doi.org/10.1063/1.3193715.
[30] R. Kempers, A. Robinson, Heated meter bar techniques: what you should know and why, in: The Art of Measuring in the Thermal Sciences, CRC Press, 2020.
[31] Thermal Management for Intel® Xeon® Processors, Intel. (n.d.). https://www.intel.com/content/www/us/en/support/articles/000006710/processors/intel-
xeon-processors.html (accessed January 20, 2023).
[32] How to Check and Monitor Your CPU Temperature, How to Check and Monitor Your CPU Temperature. (n.d.). https://www.avast.com/c-how-to-check-cpu-
temperature (accessed July 2, 2023).
[33] Information about Temperature for Intel® Processors, Intel. (n.d.). https://www.intel.com/content/www/us/en/support/articles/000005597/processors.html
(accessed May 30, 2023).

15