The Tier-1 supercomputer, named Sofia and managed by the Vrije Universiteit Brussel (VUB) for the next six years at the Nexus Datacentre in Zellik, will support scientists from a wide range of disciplines in their research. In a time of climate change, climate scientists use supercomputers with immense computing power to design, test and apply their models to all possible climate scenarios.
“The Tier-1 supercomputer is a blessing for research in weather and climate,” says VUB professor and climate scientist Lesley De Cruz. “In the past, I already used previous generations of Flemish supercomputers to run high-resolution climate simulations, allowing me to ‘zoom in’ on a specific region and obtain a more accurate assessment of local phenomena such as increases in heatwaves and extreme rainfall.”
“Our research has been accelerating for several years thanks to artificial intelligence (AI),” De Cruz says. “You often hear that AI is bad for the climate, but that refers to the careless use of large language models (LLMs). In meteorology, AI actually saves energy by replacing complex, energy-hungry calculations with specialised, fast AI models. These models first need to be trained, and for that we need powerful GPUs, such as those in the new Tier-1 supercomputer.”
GPUs (Graphics Processing Units) are specialised processors used in supercomputers to carry out huge numbers of calculations simultaneously. Thanks to their parallel architecture, they are particularly suited to applications such as artificial intelligence, simulations and data analysis. In modern supercomputers, GPUs usually work together with CPUs: the CPU organises the tasks, while the GPU handles the heavy computational workload. With their high computing power per watt and a well-developed software ecosystem, GPUs are an essential part of the current generation of supercomputers, especially for AI applications.
“We’re looking forward to training our next ultra-fast prediction model for local climate extremes (heat and rainfall) on the new supercomputer,” De Cruz adds. “The AI rainfall model learns from past radar and satellite images how rain will intensify or weaken in the coming hours. It learns in the same way as other AI tools such as DALL-E (DALL-E is an AI model that generates original images based on text descriptions, FSt), using a so-called diffusion model. But instead of pictures of kittens, it generates the most likely future scenarios for rainfall in the next few hours. Such a model gives crisis managers the chance to carry out last-minute evacuations when a deluge threatens, as happened in the summer of 2021 in the Vesdre Valley.”
Supercomputer Sofia will be used by researchers at all Flemish universities and universities of applied sciences, knowledge institutions and companies, and is part of the Flemish Supercomputer Centre (VSC). With its sustainable design and impressive computing power, the installation marks a major new step forward for research and innovation in Flanders and a milestone in the Flemish landscape of AI and HPC (High-Performance Computing) systems.
Sofia was funded by the Research Foundation – Flanders (FWO), which invested €8.6 million in the purchase of the new Flemish supercomputer thanks to support from the Flemish government.
VUB professor and climate scientist Lesley De Cruz