AI, supercomputers and how we balance water usage in the Cloud
With all of the market buzz around AI and ChatGPT 4, I thought i'd write this post to give a bit of insight into the impact on datacentre usage in terms of water (my personal area of interest).
As most people are now aware, data centers are the backbone of our digital world, processing, storing, and managing vast amounts of data that we rely on every day. The modern world is demanding increased compute power and as this need for data storage and processing grows, so does the need for efficient cooling systems to manage the heat generated by data center equipment. Traditional cooling methods, such as air conditioning, consume massive amounts of energy and water, posing environmental and financial concerns.
There is a huge amount of potential to utilise this heat productively for society and I was pleased to see the news reports this week of (29) Deep Green: Overview | LinkedIn and how they've productionised the use of waste heat to power swimming pools. This is the kind of innovation that the planet needs more of!
At Microsoft our Azure cloud uses two innovative cooling technologies, adiabatic cooling and liquid immersion, which are revolutionising the way our data centers manage heat and water usage. I thought a brief primer on these technologies and their impact on water usage in data centers, would highlight their potential to create more sustainable, efficient, and eco-friendly data centers in the future.
Understanding Data Center Cooling
Data center cooling is essential for ensuring optimal performance and longevity of the electronic equipment housed within. The primary goal is to maintain a stable temperature and humidity level, as excessive heat can cause equipment failure and data loss. Traditional cooling methods rely on air conditioning units and water chillers, which consume significant amounts of energy and water.
Adiabatic Cooling:
Adiabatic cooling, also known as evaporative cooling, is an environmentally friendly alternative to traditional cooling systems. This method relies on the natural process of evaporation to cool the air, consuming less energy and water in the process.
The basic principle behind adiabatic cooling is that when water evaporates, it absorbs heat from its surroundings, lowering the temperature. Adiabatic cooling systems utilize this principle by passing warm air through a wet medium, such as a specialized cooling pad, where the air is cooled as the water evaporates. The cooled air is then circulated throughout the data center to maintain the desired temperature.
Liquid Immersion Cooling:
Liquid immersion cooling is another innovative method for managing heat in data centers. Instead of using air as the primary medium for heat transfer, this technology submerges electronic equipment in a dielectric coolant liquid. The liquid absorbs the heat generated by the equipment and is then cooled through a heat exchanger or an external cooling system.
There are two main types of liquid immersion cooling: single-phase and two-phase. Single-phase immersion cooling uses a dielectric liquid with a high heat capacity and low viscosity, while two-phase immersion cooling involves a coolant that undergoes phase change, such as boiling and condensing, to absorb and dissipate heat more efficiently.
You can see both of these technologies at work in our Microsoft Datacenters using this interactive tour.
领英推荐
As the use of AI explodes then our ability to balance the ecological equation of water and power consumption becomes even more important.
To build the supercomputer that powers OpenAI’s projects, Microsoft linked together thousands of Nvidia graphics processing units (GPUs) on our Azure cloud computing platform. In turn, this allowed OpenAI to train increasingly powerful models and “unlocked the AI capabilities” of tools like ChatGPT and Bing.
The scale of the cloud-computing infrastructure OpenAI needed to train its models has been unprecedented – exponentially larger clusters of networked GPUs than anyone in the industry has tried to build previously. You can read about this in detail here.
So, what's in it for Water?
As the UK water industry increasingly turns to the Cloud to help transform their capabilities, it is imperative that we also look to track the carbon and water usage of their Cloud footprint. We all have a responsibility to use and consume less water and by utilising these innovative technologies in the Azure cloud, Microsoft can play its part in this effort.
Adiabatic cooling offers several advantages over traditional cooling methods in terms of water usage. Firstly, it consumes significantly less water than traditional air conditioning systems. While water is used in the evaporation process, the amount required is much lower than that needed for cooling towers in conventional systems.
Additionally, adiabatic cooling systems can take advantage of ambient air and humidity to reduce the need for water. In climates with high humidity, the air already contains moisture, which can help facilitate the evaporation process without the need for additional water. Furthermore, adiabatic cooling systems can be designed to recycle and reuse water, further minimising water consumption.
Liquid immersion cooling can also significantly reduce water usage in data centers. Because the cooling process relies on direct contact between the coolant liquid and the equipment, heat transfer is more efficient than air-based cooling methods. This results in less energy consumption, which in turn reduces the need for water-intensive cooling solutions.
Many liquid immersion cooling systems are closed-loop, meaning the coolant is continuously circulated and reused without the need for continuous water input. As a result, these systems can operate with minimal water consumption.
Benefits of Adiabatic Cooling and Liquid Immersion Technology Beyond Water Savings
While the reduction in water consumption is an obvious significant advantage, adiabatic cooling and liquid immersion technology offer several other benefits. Both technologies consume less energy than traditional cooling methods. Adiabatic cooling relies on the natural process of evaporation, requiring less energy input, while liquid immersion cooling's efficient heat transfer reduces energy consumption. With reduced energy and water consumption, data centers can experience lower operating costs. These savings can be substantial over time, making these technologies financially attractive and the Azure cloud more efficient and effective for our customers.
By providing more efficient cooling, these technologies also enable improved performance and reliability of the servers themselves, reducing the risk of thermal-related failures and extending the life of the hardware and by reducing energy and water consumption, adiabatic cooling and liquid immersion technology contribute to a reduced environmental impact.
Microsoft are investing heavily in these technologies as they can play a vital role in helping data centers achieve sustainability goals and reduce their carbon footprint. Our responsibility is to power every person and organisation on the planet to achieve more and this is part of the way that we do that whilst being responsible for the environment.
Great article matey. Project Natick fascinated me and I know it was a pilot so I'm not sure of it's future track, I'm sure you will, but I often wondered if it did become popular and it led to thousands of Naticks, that could have an adverse effect on sea temperature right?
Great insight Lewis thanks for posting
Data & AI Industry Lead for Retail and Consumer Goods
2 年Mark Bjornsgaard - Inspiring the way!