Towards a Sustainable High Performance Computing
As technologies like the Internet of Things (IoT), Artificial Intelligence (AI) and 3D imaging advance, the amount and complexity of data that needs to be processed is exponentially growing. In order to support these technologies we need fast, powerful and reliable computing resources and this is where Sustainable High Performance Computing (HPC) shines – HPC is considered fundamental to groundbreaking scientific discoveries and game-changing innovations.
What is High Performance Computing and why is it important?
As explained in previous articles, High Performance Computing (HPC) – often referred to as supercomputing – is the ability to parallelize the process of large amounts of data and perform complex calculations at high speeds. HPC is used in many verticals and across the fields of Artificial Intelligence, rendering, robotics, gaming, metaverse , weather forecasting etc. The compute power delivered by HPC clusters far exceeds that provided by a regular home computer and allows processing of big data akin to that required to build AI models used in machine learning or deep learning.
The amount of information and data available in today’s market pushes data collection and processing boundaries. In many cases, HPC is the only tool that makes sense given the vast amount of data that needs to be processed and analysed. Today’s world is data-driven and accelerating digital transformation is a critical process for most companies and governments (another great HPC use case, as it fuels process automation).
Anyone who needs to synthesise a huge amount of data into useful information and insights is a use case for HPC. High performance computing is a very important tool for research and education in sciences and engineering. It is essential for computational fluid dynamics, thermal examinations or stress and strain analysis, as well as for artificial intelligence and global warming dynamic modelling. The media and entertainment industry relies on HPC for film editing, rendering, CGI (computer generated imagery) and streaming. In the financial services industry, HPC shines when it comes to tracking market trends (applicable for both stock and crypto markets) and trading automation. As building in-house HPC resources is extremely capital intensive, HPC as a Service is a more prevalent option.
HPC sustainability impact
High performance computing is not just efficient in data processing, it is also efficient in converting electricity into heat, as is any other machine or device. In order to keep the HPC equipment performance at a high level a proper cooling system must be put in place. This means additional energy consumption to power a cooling system. Just to put things into perspective, data centres consume 200 TWh each year worldwide, where 38% (76 TWh) is estimated to go toward cooling processes.
HPC is approximately 8 times more energy-intensive than general-purpose computing and this is just the tip of the iceberg. In most data centres worldwide, over one-third of electricity consumption can be attributed to inefficient air-cooled systems in which heat is wasted and released into the atmosphere. Studies show: “It is estimated that data centres have a yearly carbon footprint of 100 megatons of CO2eq related to the production of electricity they are relying on, similar to the entire American commercial aviation sector. Unsurprisingly, this will only increase […] by 2- to 9- fold in the next decade.” The challenge is not only about energy consumption but also about energy sources, or energy mix used to power data centres, all of which impacts the carbon footprint of any company, regardless of how it consumes HPC (in house, or as a service).
The need for identifying and objectively assessing the environmental impacts of HPC is paramount. How do we put numbers to the HPC sustainability case? How do we systematically approach the problem and analyse the environmental impact of HPC operations? What are the different environmental impacts of HPC? Answering these questions is a complex task, especially due to the broad influencing variables in the HPC industry such as regional energy production mix, operating models and cooling solutions, all of which influence the final amount of CO2 emissions. One way to deal with this complexity is to narrow down the focus to a specific product or service – a “functional unit” – and to apply a scientific method called Life Cycle Inventory and Assessment (LCIA) that can be complemented by an evaluation of the infrastructure Power Usage Effectiveness (PUE).
Life Cycle Inventory Analysis (LCIA) or Life Cycle Assessment (LCA), are scientific and well documented procedures (ISO 14040 / 14044) accounting for all the environmental impacts, from cradle to grave, during the entire product’s life cycle. Inventory Analysis is a systematic, objective, stepwise procedure to identify and quantify all energy and raw material flows required and their resulting environmental impacts such as atmospheric emissions, water borne emissions, solid wastes, and other substances released to the ecosystems throughout the entire life cycle of a product or service. In other words, LCIA is a process of data collection and calculations intended to quantify the inputs and outputs required and released by a product or a service. These inputs and outputs may include natural resources used, as well as released to air, water, or land. Previous LCIAs performed on data centres results are unanimous: the operation phase, including the server use, maintenance and cooling energy consumption, is the most important and most dominant (80% – 90% of overall impacts) when compared to the production phase. However, there are very few, if any, public LCIAs publications from the major players in the HPC industry.
On the other hand, Power Usage Effectiveness (PUE) is a simple measurement of the ratio between the total power consumption of a data centre and the amount of energy consumed to run the IT equipment operation . Without any heat recovery, this ratio is always greater than 1. The larger the PUE number the less efficient.
Sustainability is more than just energy usage
So what have we learned so far? High performance computing is pivotal for the development and advancement of many industries and disciplines. At the same time, HPC is extremely power hungry and has a huge carbon footprint and significant environmental impacts. Given the importance of HPC for our modern societies, it is not an option to compromise efficiency in order to reduce environmental impacts. These impacts should be analysed in three categories:
- Scope One Emissions: energy production
- Scope Two Emissions: energy usage
- Scope Three Emissions: value chain – in the case of DeepSquare, hardware, hardware repurposing and, eventually, disposal of redundant hardware
As Oliver Peckham explained in his article, “HPC’s footprint doesn’t stop at energy use (“scope two” emissions). “Scope three” emissions, which include embedded carbon emissions and environmental impacts from along the value chain, are also high for the sector.” This gets even more important when we consider the relatively short lifespan of the equipment needed for HPC resulting from the fast-moving technology. Add to this equation that “scope two” emissions largely depend on “scope one” emissions.
We must take a holistic and systemic approach to understand the issues associated with HPC, accepting that sustainability concerns don’t start and end with electricity consumption rather also include social and economic aspects. Research has shown that the emissions associated with the manufacture of the equipment are negligible compared to the emissions induced by the consumption of electricity during the operational phase. However, the short lifespan of the equipment has important impacts on other environmental areas such as the depletion of natural resources (water shortage, critical raw materials, etc.) and the toxicity of ecosystems, to name a few. By relying on holistic and systematic approaches, DeepSquare has been able to design a more sustainable system by considering the multifunctionality of the service as both a HPC provider and a heat provider. Electricity consumption and sustainability are not synonymous. We can be big electricity consumers but consume it in a sustainable way; taking into account efficient cooling technologies combined with a reflection on where to reuse heat for other societal activities (social and economic dimensions) thus making the most of our own consumption of renewable energy.
Amongst all other environmental impacts the “cooling operation” is the most important one – of course, beside the data processing operation. Consuming additional power just to release the heat into the atmosphere is definitely a poor solution. The proper cooling system should not just be analysed by its additional energy consumption, but also by its design to do more than just suck the hot air out of the room and infuse the room with cold air. One very promising alternative is to efficiently recover and reuse heat through the process of immersion cooling, the system used by DeepSquare. Immersion cooling is the practice of submerging portions or entire systems of electronics in a liquid which is non-electrically conductive, but thermally conductive, delivering an environmentally friendly alternative to traditional air cooling.
At DeepSquare, we are using single-phase immersion cooling systems.The coolant liquid used in the system is 99% more conductive than air and we have managed to turn the otherwise wasted heat into a valuable resource by capturing heat generated by the hardware through an exchanger. Our POC (Proof of Concept) cluster in Sion is connected to the city district heating system enabling our recovered heat to be used to heat homes and domestic hot water.
As Munther Salim wrote in his article published in CIO, “Many businesses are under pressure from external stakeholders to improve sustainability and reduce their overall carbon footprint, but they need to accomplish this goal in a way that maximises business benefits”. Of course, maximising business benefits is not a limit to what one company could and should do to reduce its carbon footprint. Although some companies opt to spend more on their operations to maximise their positive impact on the environment, the decision for a company to be more sustainable does not inherently imply additional financial investment rather a new point of view.
DeepSquare is committed to do its part in contributing to the overall development and growth of the sustainable HPC industry and is currently contributing to the research in this area through a dedicated sustainability team. Our goal is to remain at the forefront of innovation in supercomputing while also making the most sustainable option be the obvious choice for businesses. In the upcoming articles in this series we will dive deeper into analysing different factors affecting the overall CO2 footprint and PUE achievements, providing insight into how DeepSquare is solving these challenges.
1- How to stop data centres from gobbling the world’s electricity: https://www.researchgate.net/publication/327596779_How_to_stop_data_centres_from_gobbling_up_the_world’s_electricity
2- Carbon footprint, the (not so) hidden cost of high performance computing https://www.bcs.org/articles-opinion-and-research/carbon-footprint-the-not-so-hidden-cost-of-high-performance-computing/
3- University, Galgotias. (2006). Life Cycle Inventory Analysis (LCIA): https://www.researchgate.net/publication/252225470_Life_Cycle_Inventory_Analysis_LCIA
4- Data Center PUE – What Does It Meanhttps://www.datacenterknowledge.com/archives/2012/03/20/data-center-pue-what-does-it-mean/
