back

“AI is at the top of our agenda. With our research, we are supporting the NRW government in transforming the state from a coal region to an AI region. Artificial intelligence is perhaps the most powerful driver of change here.”

Professor Astrid Lambrecht is Chairwoman of Forschungszentrum Jülich (FZJ), one of the largest research institutions in Europe with more than 7200 employees. The operation of supercomputers already has a long tradition at Jülich. In 1961, the Center for Applied Mathematics was founded there, which was renamed the Jülich Supercomputing Centre (JSC) in 2004. In 1967, the first IBM/360-75 high-performance computer was installed, followed by the first supercomputer CRAY X-MP six years later. This led to the founding of the Jülich Höchstleistungsrechenzentrums (HLRZ) in 1987 – the first German supercomputer center. Since then, Forschungszentrum Jülich has been one of the leading European locations for High-Performance Computing (HPC) and, with its unique computing infrastructure and technical expertise, makes an important contribution to the application of artificial intelligence (AI). In an interview with Prof. Lambrecht, we take a look at the current developments that go hand in hand with rapid progress in AI.

To begin with, you are the Chairwoman of the Board of Directors of Forschungszentrum Jülich – a huge research center in the heart of Europe, not far from Aachen. How is your institution set up, what is your goal?
Here at Jülich, we work in three areas of research that are crucial for technological independence and the future of our country: Information, energy and bioeconomy, i.e. areas that will continue to occupy people intensively for a long time to come. We have an excellent interdisciplinary research infrastructure and are represented nationally and internationally at 18 outposts. We are a founding member of the Helmholtz-Gemeinschaft and are very well networked internationally with research institutions. In Jülich, we conduct research on socially relevant topics such as digital transformation, climate protection, the energy transition and the development of a sustainable circular economy.

The demand for supercomputers and data centers is currently booming. According to Handelsblatt, 100 billion dollars will flow into such infrastructures this year alone. What is the reason behind the growing demand and how do you assess this development?
Large supercomputer infrastructures are essential for today’s research, especially for the development and use of artificial intelligence, simulations and data analysis. It is now important to have systems that are optimally suited to the increasingly complex applications and problems that computers are expected to solve and that are particularly efficient, energy-saving and sustainable. This has long been a central concern for us. At Jülich, we have been working with and on AI for over ten years. We will soon be installing the first European exascale computer, JUPITER. JUPITER will significantly expand the boundaries of scientific simulations in Europe and considerably increase the possibilities for training large AI models. However, our focus is not only on infrastructure, but we are pursuing a holistic research approach that encompasses three areas: Simulation and data science of high-performance computing, brain research and research on the bio- and nanoelectronics-based information technologies of the future.

What does a supercomputer like JUPITER look like and how powerful is this computing machine?
Supercomputers, such as JUPITER, are practically all built as parallel computers today. The JUPITER booster module, which will be installed at the end of this year, consists of around 6000 individual computers that are connected to each other via a high-performance network. Visually, this is almost unspectacular: the computers, also known as computing nodes, are housed in so-called racks, which are set up in several rows next to each other – similar images are perhaps also familiar from large data and server centers for the Internet.

JUPITER will look similar, but will be housed in a modular container building. The idea was born out of necessity, because the schedule and costs have changed so drastically due to global political events that we had to rethink quickly. This is becoming more and more of an advantage, as we are very flexible for future adaptations thanks to the modular design. The facility will cover an area of around half a soccer pitch and consist of around 50 containers, which will be manufactured and supplied by the IT company Eviden, which is part of the Atos Group.

What ultimately matters is performance: JUPITER will be the first exascale-class supercomputer in Europe. It will be the first computer in Europe to break the barrier of one trillion computing operations per second – a “1” with 18 zeros. For comparison: at least 1 million modern smartphones would be needed to achieve this level of performance. For simple 8-bit arithmetic operations, which are typically used to train AI models, JUPITER even manages more than 70 trillion arithmetic operations per second – JUPITER’s processors, which come from NVIDIA, are very efficient here. This will make JUPITER one of the most powerful AI computers in the world. The plan is to use it to train large AI models not only in Germany, but also across Europe.

With “JUWELS”, JSC already operates one of the most powerful supercomputers in Europe. How does “JUPITER” differ from its predecessor?
Yes, that’s right, JUWELS was the leader in Europe when it was inaugurated in 2020 and is still one of the best European supercomputers. It can be considered the predecessor of JUPITER. Like JUPITER, JUWELS has a modular structure. The booster module is designed for highly parallel, computationally intensive tasks, while another cluster module is more tailored to data-intensive, universal tasks. I won’t go into too much technical detail. But this structure allows us to use and allocate computing resources flexibly and efficiently for various scientific and industrial applications. JUPITER will have a similar structure, but its computing power will be significantly higher – by a factor of more than 20. This opens up completely new possibilities, especially for computationally intensive applications such as the training of AI models, which requires a great deal of computing time.

With “JUNIQ”, Forschungszentrum Jülich also operates the first open infrastructure for quantum computing in Europe. What goals are you pursuing with this?
We can see today that progress in traditional digital computers is increasingly reaching its physical limits. This can be seen, for example, in the energy consumption of today’s exascale-class supercomputers, which is significantly higher than that of previous generations of computers. JUPITER, however, will work very energy-efficiently. The first JUPITER module, which was installed this spring, is currently in first place on the Green500 list, making it officially the most energy-efficient supercomputer in the world.

New approaches such as quantum computing, which we are researching intensively at Forschungszentrum Jülich and putting into practice, have the potential to solve certain tasks much more efficiently and quickly than is possible with conventional digital computers. Put simply, quantum computers can multi-task. Thanks to the special properties of their quantum bits, they can perform several calculations simultaneously. For example, a conventional supercomputer would search for the best active ingredient for a drug among thousands of possibilities by trying them all out one after the other. A quantum computer would test them all at the same time.

However, we have not yet reached the point where we can utilize these computing advantages of quantum computers over binary computers. There are different approaches and types of quantum computers, each of which is suitable for different types of tasks. I don’t want to go into too much detail here either. Through our JUNIQ quantum computing infrastructure, we already offer science and industry access to state-of-the-art quantum computers of various types with different levels of technological maturity.

Who can use JUPITER, including external researchers?
In general, computing time on the Jülich supercomputers and also on our quantum computing infrastructure JUNIQ is available to all researchers in Germany and Europe. There is access to the supercomputers via regular calls by the Gauss Centre for Supercomputing (GCS), which unites the three federal supercomputing centers, via the Fenix Research Infrastructure or, in future, via the EuroHPC Joint Undertaking. Independent peer review procedures will decide which projects are awarded the contract. As part of the JUPITER Research and Early Access Program (JUREAP), scientists can also access JUPITER at a very early stage in order to optimize their codes in collaboration with our experts at the Jülich Supercomputing Centre. External researchers and scientists from Jülich were able to apply with their projects at the beginning of the year and, if they are successful, work with JUPITER until next year. JUPITER will then go into general user operation.

And what about companies? Can they also benefit from JUPITER with their AI projects?
As far as possible, we would also like to make our computing capacities available to industry. The AI Service Center WestAI, to which Forschungszentrum Jülich belongs, plays an important role here. This is a BMBF-funded consortium of NRW research institutions that supports companies in the development and introduction of innovative applications with AI. As you may know, Fraunhofer is also a partner. The focus is primarily on SMEs, but also on researchers from the academic environment. The offer is extensive: it ranges from AI consulting to providing the necessary hardware and training in specific areas. For example, experts from Forschungszentrum Jülich recently developed an AI examiner for exam preparation together with a publishing house. As part of a joint research project, various language models are now to be tested and integrated into the publisher’s training programs. Researchers and companies can apply to us for 10,000 GPU-h with very little effort, and thanks to the current funding, even free of charge at the moment. The application process for research institutions is described on our website, and companies can seek advice if they are interested. WestAI picks up companies that do little or nothing with AI yet, but have recognized the potential and want to learn. And the offer is to be expanded so that more joint projects between science and industry can be created, for example commercial computing time offers with JUPITER and our other supercomputers such as JUWELS.

Looking to the future: How do you see the use and operation of supercomputers in five or ten years’ time?
AI is at the top of our agenda. With our research, we are supporting the NRW government in transforming the state from a coal region to an AI region. Artificial intelligence is perhaps the most powerful driver of change here, not only in science, but for all areas of society.

We are therefore working in all areas that will determine the future of machine computing: in addition to supercomputing, we are also working on quantum and neuromorphic computing, AI and machine learning. Our goal is to offer the most powerful infrastructure in Europe, linking all domains of computing so that different user groups from science and industry can learn, grow and benefit from each other. We want to build a strong European AI community with the guiding principle: only if everyone works together can Future IT become a real innovation booster for the country and help to successfully meet global challenges. I am convinced of this.