Executive Summary
- Our editorial theme this month is cooling innovations, so this is our introductory article with the lowdown on cooling solutions in the data centre industry. Throughout the month, we will be exploring the latest cooling innovations, trends, dive into expert insights and opinions on how cooling will impact efficiency and data centre growth.
- Air cooling is the use of air conditioning units to circulate air around racks to manage thermal dynamics; liquid cooling is the use of fluids to directly cool the hottest components within servers.
- Chemistry 101: Liquid is far more efficient at heat management than air because of its dense molecules tightly packed together. Water is up to 20 times better than air at transferring heat energy.
We’re getting a bit more sunshine, some unpredictably cold blasts of wind, but the weather is warming up (and so are our data centres). As we’re all defrosting from the winter and entering Spring, this April, we’re here to talk about all things cooling: innovations, trends, unique perspectives – you name it.
To start with, let’s go back to basics and look at the different types of cooling and the best use cases for them.
Air Cooling – the OG
The longest-established traditional method of cooling down server equipment, air cooling, is the process of using air conditioning units, fans and systems to remove heat from IT equipment and circulating cold air to keep the technology working optimally.
The layout is structured where the racks are separated by hot air and cold air corridors. Air enters the cold corridor, which is typically pumped through raised floors or air ducts, and passes through the equipment to manage the thermal activity. That same air is sent to the back of the cabinet, emerges warm as it carries away the heat into the hot corridor and back into the air tubes, to cool and the cycle continues.
There are two types of air conditioning solutions
- Server-Level Air Cooling – Fans are implemented inside the server to cool components that produce the most heat. This would be the CPU and GPU. Typically, this is for data centres that consume up to 10kW of power per rack.
- Rack-Level Air Cooling – Rather than being inside the server, the fans are at the back of the rack to create a fan wall. It’s easier to monitor and smart control these fans via this method. This is optimal for facilities consuming 18 to 20 kW per rack.
Air cooling alone is beneficial for low-to-medium density facilities such as small enterprises, smaller data centres, traditional IT businesses and edge data centres. A low-to-medium density means that the facilities consume less than 20kW per rack, so employing high-performance cooling solutions wouldn’t be necessary.
As the world’s need for more computing power has driven the growth of data centres, air cooling can’t keep up, so leaders have been faced with investing in unique cooling solutions that can handle the full ebb and flow of 24/7 operations and adapt in real-time to be energy efficient.
The old saying that they would never have liquids in a data centre is no longer true, as the world turns to an array of liquid and hybrid cooling solutions to solve their thermal management capacity and efficiency issues.
Liquid Cooling
Liquid cooling is the use of delivering liquid coolant to the CPU or GPU in the server directly. The liquid operates through a pipe network, absorbing the heat and carrying it away from the servers and equipment. There are a few different types of liquid cooling, and we’ll dive into that in just a moment, but liquid cooling is most optimal for HPC (high-performance computing) data centres that have workloads exceeding 35-50 kW per rack. In these cases, this will be high-density AI, machine learning and GPU-intensive workloads. These higher density facilities also need to lower their PUE and become more energy efficient, which factors heavily into the type of cooling solutions implemented.
The following are the primary liquid cooling methods:
Direct-to-chip (D2C) Cooling
Direct-to-chip cooling circulates liquid coolant through cold plates mounted directly on processors or GPUs. As the name suggests, the fluid is going directly to the hottest components and removing the heat as efficiently as possible. This method is really effective for high-performance processors and specialised computing equipment.
Immersion Cooling
This method submerges IT equipment in a non-conductive dielectric fluid, which transfers heat away from those hot components. The key types are as follows:
Single-phase immersion cooling, in which the coolant remains in a liquid state as it transfers heat away. A Cooling Distribution Unit pumps warm fluid to a heat exchanger to remove heat, before returning the fluid to the tank to do it all over again.
Two-phase immersion cooling is when heat transfers to a dielectric fluid with a low boiling point, which evaporates, rises and hits a condenser coil to cool down enough to transfer back into a liquid again, before dropping back into the tank.
Why is liquid cooling more efficient than air cooling?
I’m taking you on a chemistry tangent, so hold onto your seats. If we removed the proximity to the components and returned to chemistry 101, we find that liquids transfer heat a hell of a lot better than air. In fact, it is often at least 4 times as good, with water being up to 20 times better than air when it comes to absorbing heat, depending on the application and context.
To understand this a bit more, let’s delve into the terminology around conductivity. Conductivity is the ability of a material to transfer energy through contact with another material. Taking it a step further, thermal conductivity is the ability to transfer heat. Liquids transfer heat more efficiently than air because tightly packed molecules move heat faster. Molecules in gases such as air are more scattered around, making it harder for energy to transfer between them.
Stay tuned this month as we delve more into cooling innovations, trends and insights this month from us at Data Centre Insight and guest blogs from industry leaders.
