HAICORE: More than just computing power
Scientific experiments often generate vast volumes of data, sometimes reaching hundreds of terabytes. This can happen, for example, when researchers use high-throughput methods to test thousands of substances for their potential to treat diseases; or when they compare healthy and diseased tissues, analyzing gene activity across different types of cells. Interpreting these massive data sets and deriving meaningful insights requires immense computing power.
“Our role is to provide researchers from other institutions with computing capacity – and with tools to analyze their data using artificial intelligence,” says Navid Afkhami, Project Leader in charge of implementating HAICORE at the Max Delbrück Center. Working with a team from Corporate IT, Afkhami has set up a new high-performance computing system featuring eight work nodes, each equipped with NVIDIA H100 graphics processing units (GPUs.) These GPUs form the basis of machine learning models that, for example, can predict how biological systems behave under certain conditions.
“With HAICORE, we can offer researchers in the Helmholtz Foundation Model Initiative (HFMI) direct access to up to 50,000 GPU hours,” says Professor Dagmar Kainmüller, Head of the Integrative Imaging Data Sciences research group and a key force behind the project at the Max Delbrück Center. “This substantially lowers the barrier for developing new large-scale AI models.”
Closing the gap
Currently, eight HFMI groups are using the new HAICORE platform at the Max Delbrück Center as part of a pilot project. HFMI supports research that leverages machine learning to build foundational AI models. According to Afkhami, this work has high public value: “Some of these projects are focused on developing new cancer therapies or tools for personalized medicine, enabling treatments and drugs to be tailored more precisely to individual patients.” But the benefits extend beyond medical research. “We’re also supporting a project that analyzes climate data to identify strategies for reducing CO₂ emissions,” he adds.
In addition to powering machine learning and carrying out simulations, the platform also allows data to be shared with other scientists. “This kind of exchange benefits everyone,” says Afkhami.
“Providing platforms like this to Helmholtz researchers helps bridge the gap between national supercomputing centers and our in-house infrastructure,” adds Karsten Häcker, head of Corporate IT at the Max Delbrück Center. “We can offer quick, low-bureaucracy support – especially for projects that aren’t yet ready for or suited to applying for national supercomputing resources.”
Text: Stefanie Reinberger
