8 November 2023
Artificial Intelligence (AI) is considered a key technology of the 21st century. The “ChatGPT Moment” has shown the potential in AI applications. The Federal Ministry of Education and Research has now published an AI Action Plan that aims to provide new impulses for the German AI ecosystem. Prof. Thomas Lippert from Forschungszentrum Jülich was invited as an expert and provided insight into the German AI computing infrastructure, which will be significantly expanded in 2024 with the first European exascale-class supercomputer in Jülich.
The AI Action Plan identifies eleven specific areas of action that are crucial for Germany and Europe to take a leading international position in this field. The computing infrastructure plays a key role, which in the future will be further expanded specifically for AI. This is because the development of new AI models is typically extremely computing-intensive.
“We are already on a good path,” explained Prof. Thomas Lippert at the presentation of the AI Action Plan. “Since 2020, with the JUWELS supercomputer, we have one of the world’s most powerful supercomputers for AI at ForschungszentrumJülich. JUPITER will massively increase computing capacities for AI in Jülich from 2024 and will surpass JUWELS by about a factor of 45 in AI applications,” said the director of the Jülich Supercomputing Centre (JSC).
The supercomputer JUPITER is expected to break the barrier of 1 trillion computing operations per second – a “1” followed by 18 zeros – in Jülich in 2024 as the first supercomputer in Europe. For simple calculations with 8-bit, as conducted for training artificial neural networks, the computing performance is even higher by 1 to 2 orders of magnitude due to the immense number of graphics processors.
Further exascale-class computers are planned within the framework of the Gauss Centre for Supercomputing (GCS), which unites the three most powerful computing centers in Germany under one roof. In the years 2025 and 2026, additional exascale systems are to follow at the federal high-performance computing centers in Garching and Stuttgart.
Even today, applications with AI make up a large part of the projects that are computed on the JUWELS supercomputer in Jülich. “We currently count around 60 projects that use methods of AI and machine learning,” explains Thomas Lippert. Among the better-known are the OpenGPT-X language model or the image model of the German LAION initiative, which underlies the AI image generator Stable Diffusion, also a development from Germany.
“These AI models do not need to stand back from commercial models from the USA. But they have the advantage that they are published as open source and are comparatively secure in terms of data protection,” says Lippert.
Work is also being done on the access modalities to enable those interested from industry to gain even faster access to the supercomputers. “Research projects continue to come onto the systems via the proven peer-review procedures. In addition, we have now set up modified access paths that allow companies from the economy and industry to gain accelerated access. These can be used, for example, to perform pre-tests with AI models before larger investments are made,” Lippert elaborates. Further access options are being prepared. “We are open to all proposals,” emphasizes Lippert.
Head of Jülich Supercomputing Centre Director at the Institute for Advanced Simulation
