Click here or press enter for the accessibility optimised version

Cooling Innovation
in Action:
An In-Depth Look at
the COOLERCHIPS
Initiative
 Executive
Summary
The need for innovative Amid the surging demand for data centers, the need for innovative and

and sustainable data sustainable cooling solutions is greater than ever. One organization
stepping up to the challenge is the US Department of Energy, which

center cooling solutions recently launched the COOLERCHIPS initiative to spearhead innovation
in the field of data center cooling technology.

is greater than ever. This More than $40 million in grants are being awarded to 15 enterprise and

guide includes an in- academic projects that aim to reinvent the way we think about data,
energy, and the environment. These schemes are focused on

depth exploration of the reshaping conventional notions of data management, energy

consumption, and environmental impact within data centers and the

US Department of digital infrastructure ecosystem.

Energy’s COOLERCHIPS In this Tech Insights report, we take a deep dive into the
COOLERCHIPS initiative, spotlighting some of the cutting-edge

initiative. projects that are poised to revolutionize data center cooling methods
and improve the efficiency of these essential facilities.
Click here or press enter for the accessibility optimised version

How the
COOLERCHIP
S Program
Aims to
Improve Data
Center
Cooling
ARPA-E's COOLERCHIPS
program is pioneering high-
performance, energy-
efficient cooling solutions for
data centers, with the goal
of reducing the power
consumption of cooling.
By Kolawole Samuel
Adebayo
In today's digitally driven world, nothing runs don't overheat and become damaged or
without data. From streaming services and crash. This is why the US Department of
social media platforms to e-commerce and Energy's Advanced Research Project Agency
cloud computing, the relentless surge in data - Energy (ARPA-E) created
generation isn't slowing down anytime soon. the COOLERCHIPS program: "to develop
Latest estimates place the volume of data transformational, highly efficient, and reliable
generated per day at 328.77 million cooling technologies for data centers."
terabytes, with global projections for 2025
reaching 180 zettabytes. The Challenges of Keeping the
Data Center Cool
For context, 1 zettabyte equals 1 billion Storing data comes at a cost, especially in
terabytes of data – and as outstanding as terms of its environmental footprints. These
these numbers are, they are expected to footprints range from energy consumption,
swell even higher. This is unsurprising, water usage, to even the lifecycle of
especially when you consider more data has hardware. A comprehensive review of data
been created in the last few decades than in center energy estimates authored by David
mankind's history. Mytton and Masao Ashtine found that data
centers consume approximately 1% to 2% of
This exponential growth, however, comes global energy and 2% of total electricity used
with a price – a parallel surge in the demand in the US.
for data centers to store, process, and
distribute all this data. However, a significant portion of that energy
is dedicated to one critical function: cooling
And as demand for data centers grows, so the data centers. Powerful computers
too does the demand for energy to run those generate a lot of heat as they process and
data centers and to keep them cool so they store massive amounts of data, and if this
temperature is not controlled, overheating Cooling is critical to data center performance,
can occur, damaging the server equipment similar to how you'd need a fan or air
and causing crashes.
It is better to conditioning to keep yourself cool on a hot
summer day or else you could suffer from heat
When asked about the current challenges
with cooling data centers, Moshe Tanach,
improve exhaustion. This is why, on average, more than
40% of the energy consumed by these data
founder and CEO at NeuReality, told Data
Center Knowledge: "Data center cooling
efficiency of centers go into powering cooling and ventilation
systems. And whether it is a computer room air
energy is a side-effect of the compute
infrastructure's power consumption and heat
non-critical conditioning (CRAC) unit, a computer room air
handler (CRAH) unit, rear door heat exchangers,
dissipation. As much as it's a critical piece in
data centers, the larger problem to address is
components or direct expansion (DX) cooling, these cooling
systems are not cheap.
the source of the heat – the compute
infrastructure."
rather than Vladimir Galabov, director of Cloud and Data

The biggest contributor to the growing

using a less Center Research at research firm Omdia,
agreed that a big challenge for data center
amount of heat that data centers produce,
according to Tanach, are deep learning
powerful cooling is reducing energy consumption of
the air conditioners and fans cooling servers.
accelerator (DLA) systems like GPUs, Tensor
Processing Units (TPUs), and others. "And it
processor. "For over a decade the largest data center
operators have been experimenting with
is going to get worse when generative AI and ways to cool servers more efficiently. One
Vladimir Galabov, director of Cloud and Data
large language models [LLMs] widen their Center Research at Omdia avenue of experimentation has been the
deployment," he added. mechanical aspect of cooling – fans. Server
fans consume a substantial share of the
electricity a server consumes," Galabov
told Data Center Knowledge. "Eliminating
fans altogether and using liquid cooling only, Moving Away From Legacy The COOLERCHIPS Program
or using a combination of cooling Data Center Designs Berti stressed the importance of leveraging
technologies like a rear door heat exchanger Apart from energy usage, the indirect and advanced cooling solutions like air-cooled
plus direct-to-chip cooling is another avenue direct consumption of water for either chillers that can improve data center cooling
being explored. generating electricity or cooling data centers while reducing costs.
also poses a climate risk if it continues to run
"I expect that optimization around the unchecked. This was echoed by Bruno Berti, This is where the COOLERCHIPS (which
mechanical components of computing and senior vice president of product and go-to- stands for Cooling Operations Optimized for
power conversion would be two key data center market at NTT Global Data Centers, who Leaps in Energy, Reliability, and Carbon
efficiency avenues that will be explored by data told Data Center Knowledge that legacy data Hyperefficiency for Information Processing
center operators," he added. "It is better to center designs use "evaporative cooling Systems) program can help. ARPA-E
improve efficiency of non-critical components technologies that waste a lot of water." While designed the COOLERCHIPS program to
rather than using a less powerful processor." Berti admitted that these legacy technologies revolutionize data center cooling by
are "very efficient and help to get data leveraging cutting-edge technologies and
A Global Markets Insight report on the data centers cool, [they are] obviously very implementing energy-efficient strategies to
center cooling market found that in 2022, the wasteful from the perspective of water," reduce the total amount of energy needed to
market size was at $10 billion. The report also impacting climate in adverse ways, he added. cool data centers in any US location.
noted that the urgent need to reduce the
carbon footprint of data centers (which is This is why NTT Global Data Centers moved
currently at more than 1% of global energy- away from evaporative cooling technologies
related GHG emissions and 3% in the US.) is and techniques to air-cooled chillers – a
driving the implementation of energy- closed-water system that doesn't waste
efficient cooling systems. water, according to Berti.
Creating solutions to cool data
centers efficiently and reduce the
associated carbon emissions
supports the technological
breakthroughs needed to fight
climate change and secure our
clean energy future.
Jennifer Granholm, US Secretary of Energy
With funding of $42 million for the program, 15 "Creating solutions to cool data centers 10% will lead to huge savings in data center
projects located at universities, businesses, efficiently and reduce the associated carbon costs and improve efficiency, he said.
and national labs will receive grants to develop emissions supports the technological
high-performance, energy-efficient cooling breakthroughs needed to fight climate change
solutions for data centers. Some of the and secure our clean energy future," said
recipients include Nvidia, the University of US Secretary of Energy Jennifer Granholm.
California, Hewlett-Packard (HP), the University
of Arkansas, Intel Federal, and Purdue For Galabov, "Any research and funding that can
University, with funding ranging from $1.2 enable data efficiency is most welcome." But, he
million to $5 million for each recipient. added, "how successful the program will be
depends on the attractiveness of the innovations
These projects will oversee the development that get developed as a result of the funding."
of technologies like secondary cooling loop
components, cooling system software, One big concern that Galabov doesn't see
cooling systems for modular/edge data many COOLERCHIPS projects addressing is
centers, and even support facilities for testing the availability of non-toxic and affordable
the new technologies — all of which are fluids for liquid cooling. "This is an area that
designed to reduce the power used for requires development," he said. "This would
cooling to a mere 5% of the data center's be something that we all need to care about."
total energy consumption, as opposed to the
33% to 40% currently used for cooling. Because of the tremendous amount of
electricity used by data centers, even small
In turn, this reduction will lower the operational changes can make big differences, Galabov
carbon footprints of these data centers and concluded. In an industry that spends $3
contribute to environmental sustainability. trillion yearly on electricity bills, even a
reduction in electricity consumption by 5% or
Click here or press enter for the accessibility optimised version

COOLERCHIP
S Project
Takes On Data
Centers' Chip
Cooling
Challenge
Chips are getting hotter, so
we need better cooling
solutions for data centers.
Here's how the University of
Florida is using its $3.2M
COOLERCHIPS grant to
develop an innovative way to
cool chips.
By Charles O'Hay
With the increased demand for artificial functionality, and storage, and chips are
intelligence and the vast amounts of data being asked to carry more of the load.
needed to build AI services coupled with the
increasing volume of data generated by other An increasingly urgent challenge is to find a
sources, the need for sustainable and scalable new approach to cooling data centers that
data storage solutions is becoming more reaches beyond legacy thermal technologies –
urgent. However, an increase in data center one that is both energy-efficient and scalable –
capacity to fill this need is also resulting in an with the ultimate goal of enabling greater data
increase in energy consumption. And this storage in an energy-efficient context.
increase in data center energy demand is
testing the capabilities of legacy thermal One organization stepping up to the
technologies, often to their limits. challenge is the U.S. Department of Energy,
which recently launched the Advanced
Data centers are complex systems in which Research Projects Agency-Energy (ARPA-E)
multiple technologies and pieces of hardware Cooling Operations Optimized for Leaps in
interact to maintain safe and continuous operation Energy, Reliability, and Carbon
of servers. With so many systems requiring power, Hyperefficiency for Information Processing
the electrical energy used generates thermal Systems – COOLERCHIPS – initiative,
energy. As the center operates, this heat builds awarding $40 million in grants to 15
and, unless removed, can cause equipment enterprise and academic projects aimed at
failures, system shutdowns, and physical damage improving data center cooling technology.
to components. These projects represent thought leadership
that is reinventing the way we think about
Much of this increased heat can be attributed data, energy, and the environment.
to CPUs and GPUs. Each new generation of
processor seems to offer greater speed,
Each of the technologies developed is the processor by lowering its frequency or testing the capabilities of legacy thermal
expected at minimum to meet Tier throttling it. While effective in the short term, technologies, often to their limits.
III reliability levels of 99.982% uptime. The repeated throttling can have negative effects,
grants will support research, groundbreaking such as shortening the life of the component. Data centers are complex systems in which
prototypes, and scalable solutions geared to In an ideal scenario, CPUs and GPUs don't multiple technologies and pieces of hardware
reshaping the landscape of data centers so require as much power consumption and thus interact to maintain safe and continuous
they meet a sustainable standard. don't get as hot. operation of servers. With so many systems
requiring power, the electrical energy used
One recipient of a COOLERCHIPS grant is Sometimes a game-changing technology generates thermal energy. As the center
the University of Florida at Gainesville, which forces us to re-evaluate our legacy systems. operates, this heat builds and, unless removed,
is using its funding to develop a solution for The increased growth and sophistication of AI can cause equipment failures, system shutdowns,
cooling CPUs and GPUs. have spurred chip designers to create larger and physical damage to components.
and more powerful chips to manage the
Why CPUs and GPUs Are demands of large-scale language training Much of this increased heat can be attributed
Heating Up programs required by AI developers. to CPUs and GPUs. Each new generation of
Before delving deeper into the University of processor seems to offer greater speed,
Florida's COOLERCHIPS project, it's important to With the increased demand for artificial functionality, and storage, and chips are
understand why CPUs and GPUs are heating up. intelligence and the vast amounts of data being asked to carry more of the load.
needed to build AI services coupled with the
Effective operation of any processor depends increasing volume of data generated by other An increasingly urgent challenge is to find a
on temperatures remaining within designated sources, the need for sustainable and scalable new approach to cooling data centers that
thresholds. The more power a CPU or GPU data storage solutions is becoming more reaches beyond legacy thermal technologies –
uses, the hotter it becomes. urgent. However, an increase in data center one that is both energy-efficient and scalable –
capacity to fill this need is also resulting in an with the ultimate goal of enabling greater data
When a component approaches its maximum increase in energy consumption. And this storage in an energy-efficient context.
temperature, a device may attempt to cool increase in data center energy demand is
 One recipient of a COOLERCHIPS grant is
the University of Florida at Gainesville, which

Data centers, being pivotal hubs of is using its funding to develop a solution for
cooling CPUs and GPUs.

digital infrastructure, have a critical Why CPUs and GPUs Are

role to play in reducing greenhouse Heating Up
Before delving deeper into the University of

gas emissions and promoting Florida's COOLERCHIPS project, it's important

to understand why CPUs and GPUs are

energy efficiency, heating up.

Effective operation of any processor depends

Saeed Moghaddam , Project Lead of University of Florida’s COOLERCHIPS program
on temperatures remaining within designated
thresholds. The more power a CPU or GPU
One organization stepping up to the These projects represent thought leadership uses, the hotter it becomes.
challenge is the U.S. Department of Energy, that is reinventing the way we think about
which recently launched the Advanced data, energy, and the environment. When a component approaches its maximum
Research Projects Agency-Energy (ARPA-E) temperature, a device may attempt to cool
Cooling Operations Optimized for Leaps in Each of the technologies developed is the processor by lowering its frequency or
Energy, Reliability, and Carbon expected at minimum to meet Tier throttling it. While effective in the short term,
Hyperefficiency for Information Processing III reliability levels of 99.982% uptime. The repeated throttling can have negative effects,
Systems – COOLERCHIPS – initiative, grants will support research, groundbreaking such as shortening the life of the component.
awarding $40 million in grants to 15 prototypes, and scalable solutions geared to In an ideal scenario, CPUs and GPUs don't
enterprise and academic projects aimed at reshaping the landscape of data centers so require as much power consumption and thus
improving data center cooling technology. they meet a sustainable standard. don't get as hot.
Sometimes a game-changing technology Today's computer chips use something called fin Cooling Today: The Big Four
forces us to re-evaluate our legacy systems. field-effect (FinFET) transistors. The internal Cooling technologies depend largely on four
The increased growth and sophistication of AI resistance of a single FinFET transistor is low, at elements: conduction, convection, layout, and
have spurred chip designers to create larger approximately 12 milliohms, but add 80 amps of automation. Each successive advance has
and more powerful chips to manage the current, and the dissipated energy resistance represented a step forward in efficiency, but
demands of large-scale language training increases to 90w per second. Multiply this by with today's increasing data demands, these
programs required by AI developers. the number of transistors housed in CPUs and four elements can often be found working in
GPUs within a data center, and the thermal concert within a single facility. Let's look at
For example, Nvidia's A100 and AMD's M100 management challenges become clear. each in greater detail.
represent a new generation of "monster
chips." The Nvidia A100 contains 54 billion Legacy Cooling Technologies: Conduction
transistors, with a die size of 826 mm² Two Approaches Used to keep the earliest servers from
and can execute 5 petaflops of performance, Traditionally, there have been two types of overheating, conduction relies on direct
or about 20 times more than Nvidia's cooling technologies employed in data centers – surface contact to move heat from hot spots
previous generation Volta chip. The AMD and neither has focused on improving the chips to cooler areas where it can be safely
M100 delivers 25.6 million transistors on a die themselves but rather on managing the dissipated. Heat spreaders allowed thermal
size of 750 mm² and is also capable of environment in which they operate. The first energy to be moved away from sensitive
performing both CPU and GPU functions. The approach is localized within the server components, but the technology's
new architecture is the first GPU to break 10 infrastructure and works by moving heat away capabilities were limited. Spreaders were
TFLOPS, offering up to 11.5 TFLOPS of peak from crowded components to a place where it quickly replaced by heat sinks, which remain
FP64 throughput. Cooling these chips can be dissipated safely. The second type of an industry standard.
presents new thermal challenges for legacy cooling technology is located below the center's
cooling technologies. floor and serves to maintain the ambient Typically, a heat sink is mounted directly to
temperature, using air circulation and the heat-producing surface by means of a
convection to reduce heat stress on all the face plate. Designs have evolved to maximize
equipment within the facility. surface area and boost efficiency. To smooth
the manufacturing process, face plates are Variable-flow fans have also proven More recently, automation has entered the
generally made from die-cast aluminum. The successful in adjusting airflow during heat frame, allowing a finer level of temperature
addition of a copper center to the base plate surges from increased demand. control and introducing the possibility of fully
increases conductive efficiency, as copper autonomous data centers, where temperatures
possesses approximately 40% greater Heat pipes, another feature of convection are continually self-monitored and regulated.
conductivity than aluminum. systems, have also received upgrades to Automation also allows servers to rest
enhance efficiency. Common heat pipes components that are less in demand, and to
Convection feature a copper enclosure, sintered copper use that energy to power components in
When conduction methods could no longer wick, and a cooling fluid. Incorporated within higher-demand areas of the facility. Automated
support the increasing power demands, a the heat sink base, these pipes directly systems make use of heat sensors and cooling
second heat-removal method was required. contact the CPU, directing heat toward the fans to direct and control airflow where and
Built into the architecture of the data center, exhaust fins in the heat sink. when it's needed. Power capping technologies
convection methods are more efficient than have also allowed for less energy waste
conduction. Using the directed flow of air or Layout without compromising performance.
liquid to provide the desired cooling, convection Placement matters, especially when you're dealing
systems are able companions to the conduction with heat-generating electrical components. As Limitations of Legacy Cooling
systems already in place. Conduction features power demand exceeded the abilities of both Technology
such as heat sinks gather heat from electrical conduction and convection technologies, engineers While each of the aforementioned strategies
components, while convection moves that heat were tasked with a new challenge – to use the generated increased cooling capacity, each
away from the servers. layout of the center itself to facilitate cooling. too had limits. Use of these technologies in
concert led to changes in the way data
Advances in convection technology have led Successful approaches include removal of centers were conceived, designed, and built.
to changes in fan design, including obstructions to airflow, design adaptations to enhance Features such as raised floors, hot and cold
innovations in fan depth, blade architecture, and control airflow, and the use of symmetrical aisles, and containment became common.
and building materials to better control configurations to balance airflow within the facility.
airflow and maximize cooling capacity.
Prior to raised floors, the computer room air
conditioning (CRAC) system simply blasted large Looking Ahead: COOLERCHIPS'
volumes of chilled air into the space. Air distribution,
however, was less than ideal. To address this University of Florida Project
challenge, raised floors were introduced to provide
sub-floor cooling. Solid tiles were swapped out for According to the ASHRAE Equipment Thermal The new technology allows for significant
perforated replacements, further improving air Guidelines for Data Processing Environments, future growth in processor power, rejects
exchange and supporting more even cooling. temperatures within data facilities should be heat directly to the ambient air external to the
between 18-22°C (recommended), with data center, and would facilitate adoption
Server configurations also changed to a hot aisle- 5-25°C being the allowable limits. Staying within existing data center infrastructure with
cold aisle system in which servers were arranged in within these parameters is challenging, a primary liquid cooling loop.
parallel rows with hot air exhausts and cold air especially as demand increases for large-
intakes facing each other at each end. Such scale language training programs for AI. The challenge, according to Saeed
configurations promote airflow and increase cooling Meeting the cooling demands of today's Moghaddam, William Powers Professor of
efficiency. But even this, combined with raised hotter chips while remaining sensitive to the Mechanical and Aerospace Engineering at the
floors, remained insufficient to meet demand. A global environment remains a technological University of Florida and the project lead of
new approach emerged: containment cooling, and environmental hurdle. the university's COOLERCHIPS program, is
sequestering hot air from cold and creating a that significant energy is used to cool data
system to strictly manage airflow streams. What is being done to address the challenge? center servers, accounting for up to 40% of
Containment successfully improved cooling The University of Florida at Gainesville is the IT power consumption. "This energy is
efficiency, reduced cooling costs, and offered data using its $3.2 million grant from the mostly used in air handling units, chillers,
center designers greater flexibility and more layout COOLERCHIPS program to develop a pumps, and cooling towers that are all
options. Containment, when used with these other disruptive thermal management solution for elements of a typical cooling system,"
systems, remains the efficiency standard in data cooling future CPU and GPU chips at Moghaddam told Data Center Knowledge.
center cooling. But all that is about to change. unprecedented heat flux and power levels in
data centers server racks.
The energy used, he explained, produces a In the University of Florida's model, the chips "When we add all power consumption associated
flow of cool air running through the racks to are cooled directly using a heat sink in which with our cooling system, we come down to 4% of
cool the servers. "Each rack at UF HiPerGator liquid is boiled. The boiled liquid is then pumped IT load compared to the current 40% of the IT
[supercomputer] dissipates ~40kW. So the outside at 70°C and can be cooled using an air- load," Moghaddam said. "Our technology will
heat intensity is very high. Hence, new cooled rooftop heat exchanger with a fan that greatly reduce the CPEX and OPEX."
technologies are needed to reduce energy uses less than 2.5% of the IT load.
associated with cooling the data centers."

Moghaddam is keenly aware of the impact

that data centers have on the environment.
"Data centers, being pivotal hubs of digital
infrastructure, have a critical role to play in
reducing greenhouse gas emissions and
promoting energy efficiency," he said.

So what makes the University of Florida

initiative a game-changer? "Because chipsets'
temperature is ~80°C and, in principle, their
heat can be released to the ambient
environment at temperatures as high as 40°C
to completely eliminate the need for power-
hungry chillers and their associated systems,"
Moghaddam said. "But, because heat goes
through so many interfaces to reach the
ambient air, a temperature difference of higher
than 80-40°C is required."
COOLERCHIPS Challenges
We spoke with Panoply Group consultant and
former Informa editor Lisa Sparks. She finds the
COOLERCHIPS initiative intriguing, particularly
because grant recipients include both
enterprise and academic entities. She cautions
that an open conversation between cooling
system researchers and chip manufacturers is Conclusion
essential in determining which technologies are In a data-hungry world, legacy cooling technologies are stretched to the limit. The DOE's
going to be most compatible. COOLERCHIPS projects represent a step toward both energy efficiency and
environmental sustainability in the world of data storage.
The global chip market is changing, with China
vastly increasing its purchase of high-power The University of Florida at Gainesville is using its grant to develop an efficient system
chips to more quickly advance its AI initiatives, that directly cools chips in a heat sink using a liquid coolant, then moves that heat to
she said. With the Biden administration where it can be safely disbursed. The system can be both integrated into existing data
considering limiting tech sales to China, it will centers and incorporated within future designs.
be interesting to see how this all plays out.
The hope is that through advanced technologies like those be developed through the
Some important questions must be tackled, COOLERCHIPS program, we will be able to meet growing global data demand in an
according to Sparks. Purchasers of any new environmentally sustainable way.
technology are likely to ask whether the
system can be retrofitted to enterprise,
colocation, and hybrid data centers; how the
new chips will affect the decommissioning of
old servers; and whether hybrid processing
power is needed to achieve these aims.
Click here or press enter for the accessibility optimised version

How a New
Two-Phase
System Aims
to
Revolutionize
Data Center
Cooling
The Hybrid Mechanical
Capillary-Drive Two-Phase
Loop seeks to make
traditional cooling a thing of
the past.
By Joe Milan
Data Centers are booming. From Virginia to However, the COOLERCHIPS initiative from
California, in rural spaces and near urban the US Department of Energy has sought to
sprawl, we build data centers to reduce address these issues by funding promising
latency and match the demands for growing and innovative technology “to reduce total
capacity. These overheads are only expected cooling energy expenditure to less than 5% of
to increase as we embrace the newest a typical data center’s IT load at any time,”
technological leap: AI. which should also reduce the CO2 footprint.

AI promises to solve the climate crisis, innovate Out of the University of Missouri, one
healthcare, and help us reconnect with our COOLERCHIPS project seeks to redefine the
past. And now, with generative AI tools like cooling landscape by making traditional
ChatGPT, the adoption of this new technology evaporative cooling a thing of the past.
will only accelerate.
The Hybrid Mechanical
Inevitably, this boom raises questions about Capillary-Drive Two-Phase
data center sustainability, particularly Loop (HTPL)
amid water shortages in places like Oregon and The Hybrid Mechanical Capillary-Drive Two-
Arizona. Much of the environmental impact Phase Loop (HTPL) is a two-phase cooling
centers on one fact: processors get hot. Hot system. Like many contemporary data
processers use a lot of energy and water, with center cooling systems, it uses a liquid, like
our current cooling technology revolving water, to cool a hot chip. The chip heats the
around evaporative cooling. In Iowa, where liquid so it changes from a fluid to a vapor.
Microsoft’s Data Centers trained OpenAI’s This ‘phase change’ allows the vapor to carry
ChatGPT, the six West Des Moines data the heat away from the chip to a place where
centers gulped 6% of the water in the district. it can cool and condense back into water.
In a traditional system, chillers evaporate
water into the air to disperse heat away from
the system and necessitate the use of fresh
water to replenish itself. But HTPL is a two-
phase closed system, meaning there isn’t a
need for large-scale thirsty chillers, noisy
rooftop evaporators, or cooling towers
constantly fed by local fresh water.

Dr Chanwoo Park, project lead of the HTPL

project at the University of Missouri, told Data
Center Knowledge that “water consumption
remains at zero throughout its operation, with
the only exceptions being maintenance events.”

According to Dr Park, HTPL utilizes several

innovative elements that remove the need for
constant water replenishment. Firstly, it features
an advanced, super-efficient heat handler in its
evaporator with a large surface area, over 150
square centimeters, for moving heat around. It’s
exceptional at managing heat – more than 300
watts per square centimeter – with its low
thermal resistance (less than 0.01 K-cm²/W)
when water is used as the cooling liquid.
The evaporator also has an efficient design Another technology that separates the HTPL models are largely being ignored. Part of the
that disperses the liquid into a super-thin from its competitors is its capillary-driven problem is that they aren’t being measured,
layer before turning it into vapor through its phase separation that allows thin-film boiling Luccioni said:
capillary heat pipes. It is a proven method in to work without flooding the boiling surface.
cooling electronics that works even better “The HTPL system distinguishes itself For instance, with ChatGPT, which was
when the liquid flows against the heat. This through its exceptional energy efficiency, queried by tens of millions of users at its peak
achieves excellent cooling results at scale; Dr surpassing emerging liquid cooling a month ago, thousands of copies of the
Park claims it can handle heat ten times technologies by a factor of 100,” Dr Park says. model are running in parallel, responding to
better than regular cooling systems. In “This amazing efficiency is mainly because of user queries in real time, all while using
addition, the HTPL system can be made how it boils the liquid in a way that uses very megawatt hours of electricity and generating
bigger or smaller as needed, so it can be little energy for pumping and has a high metric tons of carbon emissions. It’s hard to
used to cool bigger computer chips of the capacity for absorbing heat." estimate the exact quantity of emissions this
future – even if the chip is as large as 150 results in, given the secrecy and lack of
square centimeters. With its ability to be scaled up or compacted, transparency around these big LLMs [Large
the HTPL system offers many possible Language Models].
However, the HTPL’s ability to work in passive cooling applications, from concentrated solar
and active modes makes it special. “In power to unmanned aerial systems. “It is With the available data, researchers at UC
passive mode, it behaves like a particularly well-suited for applications where Riverside and UT Arlington tried to estimate
loop thermosyphon,” Dr Park said, “while in a compact, lightweight, and energy-efficient water consumption using generative AI
active mode, it operates like a pumped two- cooling system is essential,” Dr Park said. programs such as ChatGPT. Their paper,
phase loop. This flexibility allows it to switch “Making AI Less ‘Thirsty’: Uncovering and
modes as needed, ensuring reliability, The Pressing Need Addressing the Secret Water Footprint of AI
performance, and energy efficiency.” This is Dr Susha Luccioni, a researcher at AI Models,” which has yet to be peer-reviewed,
similar to modern combustion car engines incubator Hugging Face and a founding estimated that “ChatGPT needs to ‘drink’ a
that shut themselves down temporarily while member of Climate Change AI, argues that 500ml bottle of water for a simple
idling at a stop light to save energy. the environmental impacts of generative AI conversation of roughly 20-50 questions and
answers, depending on when and where According to COOLERCHIPS Director Dr Peter
ChatGPT is deployed.” de Bock, “Facilities for large data centers are
typically structures that are built for 15-20
While a simple bottle of water might not seem years of use, and technology adoption might
like much, when the researchers consider the be modest at first as there is a large installed
volume of interactions with ChatGPT and the base of existing infrastructure.” Yet, the
other forms of generative AI for 2022, they HTPL’s possibility of a 100-fold increase in
estimated that data centers used about “1.5 efficiency and our era’s current generative AI
billion cubic meters of water withdrawal in boom may make its impact felt sooner rather
the US, accounting for about 0.33% of the than later.
total US annual water withdrawal,” or roughly
double the water withdrawal of the country Yet, as Luis Colon, senior technology
of Denmark. If the boom in data centers evangelist for Fauna.com, the question
continues, the researchers suspect that remains what to do with the legacy
water consumption by data centers will equipment.
double again by 2027.
“The side-effect of replacing a lot of old iron –
These estimates make the application of the costlier, inefficient machines that run warmer,
HTPL all the more appealing. Its application in weigh more, and waste a lot of space for the
new and old data centers would significantly computing power they provide. I hope the
reduce the demand for water while reducing [COOLERCHIPS] program stresses the need
energy consumption, both major objectives of for circular practices and incentivizes proper
the COOLERCHIPS initiative. recycling since less than one-fifth of all e-
waste is appropriately managed and recycled.”
Click here or press enter for the accessibility optimised version

Is Chip
Cooling the
Answer to
Data Center
Sustainability
?
After receiving a $1.8M
COOLERCHIPS grant, Purdue
University looks to
revolutionize data center
cooling with innovative chip
cooling methods.
By Soni Brown
Data centers are notorious for their high significant investments in research,
energy consumption, with cooling equipment particularly in chip cooling. The cooling of
accounting for nearly 30% of the total power microchips – major energy-consuming
usage. Data centers with processor thermal components in data centers – alone can
designs can generate 20 to 50 MW of heat, account for 40% of overall energy usage in a
according to Hugh Hudson, a data center data center, and the demand for water as a
market manager for Daikin Applied, a cooling resource is becoming a widespread
commercial cooling company. In light of this, problem. Organizations are now prioritizing
researchers are exploring chip cooling water conservation as part of their
methods to mitigate the energy burden and sustainability efforts.
improve efficiency in data centers.
Yongsuk Choi, chief strategy and
The US Department of infrastructure officer, at Empyrion DC, a
Energy's COOLERCHIPS program recently digital infrastructure platform company,
granted Purdue University a substantial says artificial intelligence (AI) and machine
award to advance its pioneering efforts in learning (ML) applications are really "shaping
developing more efficient cooling solutions the data center industry."
for data centers. With a focus on reducing
energy usage and enhancing cooling "Such growth in high-capacity computing has
efficiency, Purdue aims to revolutionize the caused a dramatic 200% increase in rack
way chips are cooled. density to 20-30KW per rack today," Choi
told Data Center Knowledge. "This is where
The Challenge of Cooling microchip cooling comes in. Compared to
Chips in Data Centers conventional air cooling, which requires
The exponential growth of data centers in an cooling of the entire service, microchip
ever-increasing digital world has led to cooling is much more targeted and addresses
the direct source of heat at the chip level, Purdue University's microchannels filled with liquid integrated
achieving faster cooling with less energy Pioneering Chip Cooling directly within the microchip packaging. As
consumption. We foresee greater adoption of Research the chip generates heat, the liquid boils, and
microchip cooling by data center operators." Purdue University, a recipient of a $1.8 million the resulting vapors carry the heat away from
grant under the COOLERCHIPS program, is at the chip. The steam then condenses and is
COOLERCHIPS Goal: Develop the forefront of developing innovative chip recirculated into the chip, enabling passive
'Highly Efficient' Cooling cooling solutions. The university's research cooling without the need for a pump.
Technologies goes beyond cooling semiconductor
To address the challenges and explore novel microchips and packaging, aiming to optimize Direct Liquid Contact and
approaches to chip cooling, the DOE heat transfer among data center Gravity-Driven Flow
launched the Advanced Research Projects components. By enhancing thermal Purdue's innovative solution also involves
Agency-Energy (ARPA-E) Cooling Operations performance and reducing pumping power, direct liquid contact with the chip's surface,
Optimized for Leaps in Energy, Reliability, and Purdue's goal is to revolutionize the way data eliminating the need for thermal interface
Carbon Hyperefficiency for Information centers are cooled. materials. A layer of silicone on the backside
Processing Systems – or COOLERCHIPS – further enhances cooling efficiency. By
program. This program aims to drive Two-Phase Jet Impingement utilizing the gravity of the liquid, Purdue aims
advancements in cooling technology through Cooling to minimize power consumption by placing
collaboration between academia, industry, The Purdue team was looking at more the condenser at a higher position and
and government national labs. "aggressive technologies" than what's in use, allowing gravity to drive the flow into the
according to Tiwei Wei, assistant professor of chips. This approach optimizes chip cooling
In May, the COOLERCHIPS program issued mechanical engineering and the project's while reducing energy consumption.
$40 million in funding to support research in lead principal investigator.
this field, with the goal to reduce the "Our solution is basically looking at impinging the
environmental impact of data centers by One of the key research areas at Purdue data cooling with some surface enhancement and
developing "highly efficient and reliable University involves two-phase jet whipper extraction technologies," said Wei. "From
cooling technologies." impingement cooling. This approach utilizes the chip level, the cooling is directly on the chip."
Meeting the Cooling Demands However, industry veteran and founder and
of High-Performance Chips CEO of Edge Cloud Link Yuval Bachar notes
Traditional cooling methods in data centers, that there are other alternatives currently in
such as cooling at the room level, are no longer use to ease the pain points of data centers.
sufficient to meet the demands of modern high- Bachar sees harnessing the power of hydrogen
performance microchips. To effectively handle as key to data center energy issues. He's
heat generation, data centers now rely on a launched a unique data center model that Conclusion
combination of air-cooling, direct-liquid, and harnesses the power of hydrogen. His data Efficiently cooling chips in data centers
immersion cooling systems. Enhancing thermal centers would house 4,000 traditional servers is a critical factor in improving overall
performance and reducing pumping power in about 100 square meters. energy efficiency and reducing costs.
contribute to energy savings and enable data Purdue University's research supported
centers to accommodate the higher transistor "I think we can get to 5% overhead with by the COOLERCHIPS program aims to
densities of new computer chips. traditional cooling solutions without the need revolutionize chip cooling methods
for emissions cooling," said Bachar. through innovative approaches such as
By developing more efficient cooling two-phase jet impingement and direct
solutions for chips, Purdue University's But Bachar and Purdue aim for the same liquid contact.
research aligns with the goal of the result: reducing cooling costs. Purdue's
COOLERCHIPS program. efforts align with the broader objective of the By optimizing cooling efficiency, data
COOLERSHIPS program, which seeks to centers can achieve substantial savings
Adopting a new class of efficient cooling systems is create energy-efficient cooling solutions and make significant progress toward a
vital, DOE Secretary of Energy Jennifer Granholm said while reducing carbon footprint. It also seeks more sustainable future.
in a statement, because soaring temperatures and innovation that is reliable and cost-effective.
increasing heat waves are impacting data centers.
"Data centers which connect critical computing and
network infrastructure" must be kept at certain
temperatures to remain operational, she added.
Click here or press enter for the accessibility optimised version

HoMEDUCS
Project's
Unique
Approach to
Keeping
Modular Data
Centers Cool
One of 15 COOLERCHIPS
projects, HoMEDUCS is
using revolutionary cooling
technology to make modular
data centers more energy
efficient.
By Joe Milan
With the advancement of AI and 5G, For modular data centers, cooling is a
concerns with privacy, and the continuing particularly important issue because of their
expansion of the internet of things, there is tight spaces and ability to be deployed to
an ever-growing demand for edge remote places requiring liquid-based
computing. Modular data centers have been immersion cooling using energy-intense
growing in popularity to meet that demand, chillers or evaporative cooling drawing from
since they can be deployed rapidly to remote the water supply.
areas as well as supplement brick-and-mortar
data centers. In May, the U.S. Department of Energy's
Advanced Research Project Agency - Energy
However, one major issue for all data centers is (ARPA-E) announced its Cooling Operations
their energy and water consumption Optimized for Leaps in Energy, Reliability, and
(particularly for cooling, which accounts for up Carbon Hyperefficiency for Information
to 40% of data center energy use), which has Processing Systems (COOLERCHIPS)
made them the subject of front-page news. program, which funds projects to reduce the
environmental impact of data centers by
In February 2023, it was revealed that a Google developing "highly efficient and reliable
data center uses a quarter of an Oregon town's cooling technologies."
water, and according to a 2021 study from Virginia
Tech, "One-fifth of data center servers' direct The fundamental goal of the project,
water footprint comes from moderately to highly according to Dr. Peter de Bock, the program
water-stressed watersheds." Much of this water is director for COOLERCHIPS, is to find "a
used for evaporative cooling, which becomes an transformational path to more energy-
issue in the historic drought conditions efficient data centers and computing" as part
being experienced in the U.S. Southwest. of the larger hope of reducing carbon
emissions to, as U.S. Secretary of Energy
 How HoMEDUCS Improves
Modular Data Center Cooling
Rather than trying to achieve better cooling
through a single element or a simple upgrade
to prior cooling methods, the HoMEDUCS
project deploys a series of significant
improvements to the whole modular data
center cooling system, starting with the basic
principle of what really needs to be cooled.

unlike offices with comfortable temperatures

between 68-76 degrees Fahrenheit (20-24.4
degrees Celsius) – as recommended
by OSHA – computer chips tolerate far higher
temperatures (158-176 degrees Fahrenheit/
70-80 degrees Celsius). (This is a fact all of
us learn from hot laptops, sometimes causing
severe burns). Computer chips can handle
HoMEDUCS data center in operation mode (courtesy of the UC Davis HoMEDUCS team)
temperatures that, even on the hottest days,
the U.S. Southwest doesn't reach.
Jennifer M. Granholm said, "beat climate Modular Energy-efficient Directed Cooling
change and reach our clean energy future." Solutions (HoMEDUCS) project.
By focusing on this idea, HoMEDUCS Project
One funded project that shows particular Lead Dr. Vinod Narayanan says, "If you have a
promise for cooling modular data centers is computer chip that is at 80 degrees Celsius,
the University of California, Davis' Holistic even if you have an outdoor ambient that's 40
degree Celsius (104 degrees Fahrenheit) ... not compressors or chillers drastically are projected to use less than 5% of a data
that [temperature difference] can be used to increases energy savings, much in the same center's power consumption for cooling and
drive the heat away from the chip." way that a simple ceiling fan draws less use no water.
HoMEDUCS' cooling project focuses on electricity than an AC unit. The cooled fluid
extracting and dissipating the heat from the returns to the cold plate on the chip and "What makes us unique," Narayanan says, "is
chip into the ambient air, starting with the repeats the process. the combination of technologies that we are
direct liquid cooling of the chip. bringing forward that make it more efficient
If outside temperatures exceed 40 degrees and compact," the very things needed for
HoMEDUCS uses a cold plate design that Celsius (104 degrees Fahrenheit), HoMEDUCS' easily deployable modular data centers that
differs from other cold plate designs by its design incorporates Skycool's radiative cooling could operate anywhere in America where
unique fluid channel design that focuses on panels on the module's roof, which can cool power is available.
smaller scales and differing geometry that liquid below the ambient temperatures without
enhances heat transfer while reducing electricity, even on a sunny day. Cooled fluid
pressure drop, thus reducing the pumping will be stored below the module and used
power needed for the cold plate fluid during times of extreme heat.
(propylene glycol).
Currently, cooling data centers using
Once the fluid exits the cold plates, it travels to evaporative or chiller-based cooling typically
a wall of ultra-efficient heat exchangers, which consumes around 25% to 40% of their
utilize the HoMEDUCS team's innovative "pure" energy, not including the incredible amounts
counterflow heat exchanger design and are of water being consumed, which have
made from polymer (for cost reduction). The caught the ire of drought-burdened
heat then dissipates to the ambient air with communities. HoMEDUCS' three design
the aid of fans; no compressors or chillers are elements – cold plates, ultra-efficient heat
involved, unlike existing modular data center exchangers, and radiative cooling panels –
cooling designs. Using only simple pumps and and others, which have yet to be disclosed,
Impacts on Data Center Cooling "generally apply to many electronic systems
Besides UC Davis' HoMEDUCS project, there are limited by the efficiency of their cooling systems
14 other COOLERCHIPS projects with energy and could also benefit power conversion
efficiency as their goal, ranging from Nvidia's systems for solar systems [and] wind turbines."
project of applying "green refrigerants" directly to
the chip cold plate with rack manifolds with built- De Bock adds that modular data centers will
in pumps and liquid-vapor separators to the likely be the biggest benefactor of whatever
University of Maryland's "Multi-Objective technologies are developed by the
Software" that aims to provide decision support COOLERCHIPS program. "Modular data centers
for data center designers in order to develop the or edge data centers include their own building
next generation of data centers. structure and can therefore more rapidly adapt
and utilize the COOLERCHIPS technologies
Luis Colón, senior technology evangelist of developed," he says. "They also offer unique
fauna.com, who has a long history of working use cases where computing can occur close to
with private and rented data centers, says that the customer for low latency, fast
the impact of the COOLERCHIPS program will be communication between the data center and
felt significantly by "hyper-scale" data centers. the user site, and the potential of waste heat
They will be able to decrease the load on their reuse for a greenhouse, heating, drying, or
locality's energy grid, especially during the other application."
extreme temperature fluctuations in winters and
summers when the local grids are stressed the Perhaps the greatest potential benefit of
most, allowing them to be "better neighbors." projects like HoMEDUCS is a future where
data centers can be, as Colón puts it, "better
De Bock believes the COOLERCHIPS program is neighbors" by using less energy and water,
more than just a path to energy-efficient data and increasing edge computing to users
centers. The developed technologies will wherever they live.
Click here or press enter for the accessibility optimised version

Thank you for reading! Want more?

Data Center
Knowledge is a leading
online source of daily
news and analysis
about the data center
industry.

Cookies Terms Privacy POWERED BY