Computational Science at Extreme Scale

The fastest supercomputers in the world have recently reached ExaFLOPS performance. This means that they can execute 10^18 floating point operations per second. While this amazing number highlights the enormous progress, in computer technology, we should of course also ask “what is this good for?” This talk will illustrate typical supercomputer applications and two examples will be presented in more detail. The first is from the Earth sciences. Here computational models of the Earth mantle must represent the volume of almost the whole planet, that is approximately 10^12 cubic kilometers. If a resolution of 1km is desired globally, then computing a single time step of the simulation will involve solving a linear system of equations with more than 10^12 unknowns. With short time steps on the order of tenthousand years, the goal is to compute the evolution of the planet for a time span of hundreds of millions of years both forward and backward in time. In the talk, we will present some of the algorithms, programming techniques, and the supercomputer systems that are necessary to bring computations of such magnitude within reach. A second example will be the simulation of wind turbines or wind farms, as one of the critical technologies needed in our the transition to renewable energies. Here simulations do not only serve to better understand the physics of wind turbines, but they also help to improve their design, with the goal of maximizing the harvested energy. These two examples illustrate scientific research, where progress depends crucially our ability to simulate such complex systems accurately enough. Here simulations on large supercomputers help to develop new scientific insight or they can be directly used in the development of technologies and as powerful tools to optimize them.