Zum Inhalt gehen

Sustainable Computing Infrastructure for Generative AI Applications

Maik Schwalm
19. Apr. 2024


In a world characterized by an exponential increase in data volumes and a growing need for powerful computing resources, the development of sustainable computing infrastructure for High Performance Computing (HPC) is the focus of attention.

Especially, applications such as Generative Artificial Intelligence (GenAI), which require immense computing power to enable innovative solutions. In this article, we will look at the importance of sustainable computing infrastructure for such applications and how it will shape the future of HPC.

The challenges of generative AI for HPC

Generative AI applications, such as neural networks for image generation or speech synthesis, require enormous computing capacities. These applications go through complex training phases that require a large number of data processing cycles. Traditional data centers based on conventional architectures quickly reach their limits and cause high energy costs and environmental pollution.

Sustainable computing infrastructure as a solution

Sustainable computing infrastructures for HPC applications offer a promising solution to these challenges. By integrating environmentally friendly technologies and efficient computing architectures, operating costs can be lowered, and their ecological footprint can also be reduced.

Energy-efficient data centers

The development of energy-efficient data centers is a key element for sustainable HPC. This includes the use of energy-saving hardware, intelligent cooling technologies and renewable energy sources. One innovative solution is to merge data centers and renewable energy infrastructures; windCORES operates the data center directly in the tower base of a wind turbine. This integration not only draws electricity directly from renewable sources, but also exploits synergies from existing infrastructures such as buildings, power grids and fiber optic connections. This increases sustainability and thus leads to massive CO2 savings of up to 95% compared to traditional data centers.

Edge computing and decentralized architectures

The use of edge computing and decentralized computing architectures makes it possible to bring computing power closer to the end user. In addition, decentralized systems can react more flexibly to fluctuations in demand, which leads to an optimal use of resources. Another advantage is that the computing power can be optimized for sustainability depending on wind intensity. Each windCORES Edge site can currently provide up to 1 MW of IT power. Up to 2,000 wind turbines across Germany are suitable for edge computing.

AI-optimized hardware

The development of hardware specifically optimized for AI applications, such as graphics processing units (GPUs) and tensor processing units (TPUs), also contributes to improving the sustainability of HPC. These specialized chips offer significantly higher computing power per watt compared to conventional processors, which leads to a reduction in energy consumption.

Die Vorteile nachThe advantages of sustainably operated HPC infrastructurehaltig betriebener HPC-Infrastruktur 

The implementation of sustainable computing infrastructure for GenAI applications offers a number of advantages. These include:

  • Cost savings: By reducing energy costs and optimizing the use of resources, companies can significantly reduce their operating costs.
  • Environmental friendliness: The use of renewable energy sources and the integration of wind turbines help to minimize the ecological footprint of HPC applications and make a sustainable contribution.
  • Sector coupling Power-to-Byte: Targeted load shifting of IT applications in windy periods can increase economic efficiency through favorable wind power tariffs and reduce the CO2 footprint.
  • Leverage innovation: Sustainable HPC infrastructure enables researchers and developers to solve complex problems faster and more efficiently, leading to new insights and innovations.

Conclusion

The development of sustainable computing infrastructure is crucial for the economically successful future of high-performance computing, especially in the context of generative AI applications. By integrating energy efficiency, decentralized architectures and specially optimized hardware, companies and research institutions can not only reduce their operating costs, but also contribute to sustainability goals and meet regulatory requirements. The multiple use of existing renewable energy infrastructure, such as wind turbines, is an effective step towards a more sustainable digital future.

Co-authors

Dr. Jochen Thäder, Dr. Fiete Dubberke (windCORES)

About windCORES

windCORES is a brand of the company WestfalenWIND, whose core business is the planning, construction and operation of wind turbines. In 2015, WestfalenStrom GmbH was added as a separate electricity sales company that produces and markets genuine green electricity from renewable energies.
 We want to make the world greener with innovative ideas. With windCORES, an idea has grown into a brand that was awarded the German Data Center Prize in 2019.

Dr. Fiete Dubberke

CEO at WestfalenWIND IT

Dr. Fiete Dubberke is co-founder and co-developer of the windCORES concept and has been responsible for the activities of WestfalenWIND IT as Managing Director since 2018. With his background in thermal process engineering in combination with renewable energies, he supports windCORES and thus enables distributed IT infrastructure to be set up in a sustainable and future-proof manner.

LinkedIn

Further information

Authors

Maik Schwalm

Head of Sustainability | Cloud Infrastructure Services
Als Experte zum Thema Nachhaltigkeit im Bereich Cloud Infrastructure Services berate ich mit meinem Team Firmen im Kontext der digitalen Transformation mit Fokus auf Dekarbonisierung in der IT und damit zum Erreichen von Kosten- sowie Nachhaltigkeitszielen. Neben den gewaltigen Herausforderungen sehe ich aber vor allem zahlreiche Chancen für mehr Klimaschutz und Erhöhung der Lebensqualität. Durch meine internationalen Einsätze wie beispielsweise in Shanghai, China und im Silicon Valley, USA bringe ich Expertise zu globalen Anforderungen sowie unterschiedlichen Kundenerwartungen mit.

Dr. Jochen Thäder

Client Partner | Market Unit Automotive | Capgemini Deutschland
Seit mehr als 15 Jahren beschäftige ich mich mit skalierbaren Daten getriebenen, analytischen Big Data und HPC-Themen und Projekten. Aus der Forschung (CERN, Berkeley LB) bzw. dem Lehrstuhl für High Performance Computing der Goethe-Universität Frankfurt kommend habe ich mich seit dem Wechsel vor 8 Jahren in die IT-Beratung in den letzten Jahren der Daten getriebenen Entwicklung vor allem im Automotive Bereich verschrieben. Dabei stehen stets nachhaltige Innovationen für unsere Kunden und Skalierung der Daten Architekturen im Vordergrund.