Hasso-Plattner-InstitutSDG am HPI
Hasso-Plattner-InstitutDSG am HPI


Towards Sustainable Digital Technologies

Artificial intelligence, blockchain, big data analysis and ubiquitous data exchange. Digital transformation permeates our everyday lives around the world and is the key to solving global human challenges, such as climate change, poverty and economic prosperity for all. However, is digitalization ecologically sustainable? No, currently it is not – but this has to change now!

Graphic: HPI clean IT - AI Carbon
Artificial intelligence emissions Training a modern AI model can use up as much carbon as 300 roundtrip flights from SFO-NYC or the life cycle of 5 cars incl. fuel. / Strubell et al. 2019

What is the problem?

Innovative business models and the transformation of entire economic sectors are almost exclusively based on the use of AI, Big Data, Blockchain and globally interconnected data centers. The greatest potential for growth lies in two areas: digital services, through the consistent development of digital platforms and in manufacturing by leveraging the opportunities of the Internet of  things. Individual products and services can be offered at the price of mass goods, thus increasing global prosperity. However, it is often ignored that digital technologies are also increasingly
the cause of global pollution. Every digital operation consumes energy and therefore adds to the global CO2 footprint. Very soon, digitalization will become the climate problem number one.

Graphic: HPI clean IT

Why GreenIT is not enough?

In recent years, there have been attempts in the context of „GreenIT“ to solve this problem. Typically, these attempts have focused on making the production of digital devices more sustainable through the efficient use of raw materials, and advocating for „digital sobriety“. However, this response seems unlikely to be sufficient, because digital technologies in other sectors can significantly contribute to reducing greenhouse gases and waste, as well as addressing important global challenges in the areas of poverty reduction, economic participation and education.

We need clean-IT and Sustainability by Design

To solve the paradox of more from less, new algorithmic paradigms need to be put into practice. The principle of „Sustainability by Design“ needs to become the very foundation of software development. In industry and academia, solutions are currently being funded and implemented that achieve particularly precise results (e.g. in AI) or transfer particularly large amounts of data (e.g. in streaming) without taking energy costs into account. Often unnecessarily complicated programming or software design causes higher energy consumption compared to algorithms
that are more efficient. Innovative software architectures can achieve the same/slightly lower precision or data throughput, while saving enormous amounts of energy. Algorithmic efficiency therefore needs to become the leading paradigm of software development. We call this approach clean-IT.


Examples of clean-IT

Graphic: HPI clean IT - Binary Neural Networks

Binary Neural Networks

While the best AI systems train neural networks based on 32-bit algorithms, the procedure can also be carried out with „binary neural networks“ (1-bit algorithm). This drastically reduces the effort in the individual calculation steps and immediately leads to energy savings by a factor of 20. Although binary neural networks are currently about 5 % less accurate than those of AI systems of global players, the reduction can save 95 % electricity usage. With AI applications being used millions of times a day, this value increases significantly.

Graphic: HPI Clean IT - Energy Aware Computing

Energy-Aware Computing

Next-generation data centers are embracing an increasingly diverse landscape of accelerators and hardware architectures, each offering advantages for certain algorithm classes or application domains. Unfortunately, today‘s software widely ignores this degree of heterogeneity. By executing workloads on the best-suited hardware, power efficiency can be improved significantly, e.g. by a factor of 10 for weather simulation models using FPGA accelerators instead of general-purpose processors.