June 16, 2025 – by Eric Gedenk
From its long-running student support during international Student Cluster Competitions to its close collaboration with ETH Zürich students, CSCS continually seeks opportunities to develop the next generation of high-performance computing (HPC) experts wherever possible. HPC centres continue to serve as laboratories for the rapid growth of generative artificial intelligence (AI) technologies, and CSCS recently expanded its academic collaboration with ETH Zürich to prepare students for the next technological frontier.
This year, ETH Zürich began offering the course “Large-Scale AI Engineering,” taught by Dr. Arnout Devos, Academic Program Manager at ETH AI Center and Dr. Imanol Schlág, Technical Lead for Large Language Model Development within the Swiss AI Initiative at the ETH AI Center. The course not only helps students move beyond a foundational understanding of AI-related development, but it also introduces them to one of the most powerful AI systems in the world—CSCS’s “Alps” supercomputer, inaugurated in the fall of 2024.
“Alps gives us access to the largest AI-ready supercomputer hosted at a public institution,” said Schlág. “Access to such a powerful system is a rare opportunity.”
Because of an increased focus on supporting next-generation AI research at CSCS, “Alps” was designed in close collaboration with Hewlett Packard Enterprise (HPE) and NVIDIA to consist of more than 10,000 GH200 Grace Hopper 200 superchips. Combining an ARM-based Grace CPU and Hopper GPU on a single chip, the architecture is well-suited for traditional modelling and simulation, machine learning and other foundational AI applications, and for developing the large-scale generative AI models of the future—exactly the point for students in this ETH course.
“Using Alps for our university work has been incredibly valuable,” says Niklas Canova, a graduate student at ETH Zürich who took the course. “It's one thing to study AI in theory, but actually working with such powerful hardware gives you a much deeper understanding and practical skill. This course has really informed my career path; getting this kind of hands-on experience in school means I'm graduating with real-world skills that make me feel well-prepared for roles in the field.”
Schlág pointed out that many universities around the world offer students courses on high-performance computing and machine learning, but these courses are often focused on algorithm development, parallelizing applications, and other core principles of scientific research on supercomputers. Far fewer courses have integrated the new skills needed for generative AI competency.
“This course was specifically designed to live at the intersection of HPC skills,” Schlág said. “We are focused on teaching students about distributed training of large language models, but these principles can be applied to any sort of neural network being trained to generate and learn from large datasets.” In addition to lectures, students must complete six assignments and a final mini-project in which they are encouraged to merge their code with other students’ features, fostering the collaborative environment typical of the work they will see in their careers as computer scientists.
As more economic sectors seek ways to benefit from the promise of generative AI, society will need more computer scientists who have the skills necessary to both help design models tailored to individual organizations’ needs as well as optimize these large, complex neural networks to run on systems like “Alps.”
In addition to providing access to “Alps”—when running their projects, students typically had access to 8 nodes on “Alps,” providing them experience in programming in parallel on 32 state-of-the-art GH200 GPUs—CSCS engineers also went to speak to the course about the centre’s mission, history, and current infrastructure.
“Ensuring that the computational resources we provide are used effectively has always been important to CSCS,” said Dr. Fawzi Mohamed, Senior Software Engineer at CSCS and instructor in the course. “With the increasing economic and social impact of AI, where computing at scale is crucial, we need to train young scientists and engineers so they can tackle the practical challenges that come from using a world-class computing infrastructure like the one we provide with “Alps.” This goes a long way toward helping Switzerland, and open research worldwide, maintain and foster know-how in competitive, practical AI and machine learning.”
Course lead Imanol Schlág indicated that the inaugural course was a success, and the ETH Department of Computer Science plans to expand how many seats are offered moving forward. He also pointed out that despite it being a masters’ course, roughly 25 attendees were either PhD students or postdoctoral researchers. “The interest from these doctoral and postdoctoral attendees just shows that this knowledge is interesting and valuable for people who are doing research in this space, both now and in the future.”