Computing power alone is not enough

Modern supercomputers can perform more than a trillion computing operations per second. Since 1965, when the American chemist Gordon Moore (1929–2023) quite accurately predicted that the number of transistors on computer chips would double within one to two years, computing power has increased a billionfold.

Such high-performance computers (HPC) are used, for example, to simulate flow conditions in aircraft and rocket engines or to model the development of galaxies and black holes. They train neural networks, optimize artificial intelligence algorithms and make weather forecasts more reliable.

In addition to the hardware, however, the software must also keep pace with the increasingly complex demands of research and technology. “We not only need more computing power, but also more efficient programs for the increasingly complex simulations,” says Hartwig Anzt from the TUM School of Computation, Information and Technology (CIT).

Take ICON, for example: The model for weather and climate forecasts was developed by the German Meteorological Service, the Max Planck Institute for Meteorology, the Karlsruhe Institute of Technology (KIT) along with other institutions. Among other things, it provides the data for the weather forecasts at the end of the “Tagesschau” news program.

ICON calculates its forecasts of future weather developments based on a grid that maps the earth’s surface. This grid contains data such as temperature, pressure and humidity from historical and current measurements. For regions where a high forecast accuracy is required, the grid – and thus the numerical approximation – can be calculated with a very fine mesh.

However, as the spatial resolution becomes finer, the amount of data increases – and with it the computational effort. “That’s why we are focusing on speeding up the calculations,” says Hartwig Anzt. “This is important because ICON performs a large number of simulations for a single forecast, each with slightly different parameters.” According to the computer scientist, such “ensemble simulations” are complex, but reduce uncertainties in the forecasts.

The team at the TUM Campus Heilbronn has set itself a particularly ambitious goal of simulating the human heart. Within the MICROCARD project of the “European High Performance Computing Joint Undertaking” (EuroHPC JU), the aim is nothing less than to create a computer model of the human heart at the cellular level. “We know that the individual heart cells communicate with each other using electrochemical processes,” says Anzt. “But we don’t yet understand exactly how this works and which anomalies lead to cardiac arrhythmias or a heart attack.”

Until now, the computing power for simulation at the cellular level was simply not sufficient, as our heart contains several billion cells. With the help of GINKGO, this should now be possible. In the future, Anzt hopes that the heart can be represented as a “digital twin.” Diagnosed cardiac arrhythmias or other symptoms of patients can then be simulated in the model – with the aim of identifying the regions in the digital heart where the electrical conduction is disrupted in the affected individuals. This means that cardiologists know at an early stage where and how to start treatment, for example, to prevent an impending heart attack.