High-performance and GPU computing

To fully exploit modern computer architectures (such as GPUs) often requires purpose-built algorithms. To perform development and numerical simulation we maintain a local GPU cluster with 31 GPUs (more details for using the system can be found here, only accessible with the network of the University of Innsbruck).

We have accelerated a range of computer simulations using GPU computing. For example Vlasov simulation in plasma physics, Schrödinger equation in quantum mechanics, fluid flow over airfoils, and sonic boom propagation. We are further interested in exploiting specific features of modern computer hardware in order to accelerate computer simulations. For example, the improved performance of single precision computation on GPUs (especially on the cheaper consumer cards) can be exploited by mixed precision algorithms.