The DFG research unit Holistic Energy and Performance Modeling for Sustainable Computing (Mod4Comp) puts its central focus on the advancement of performance and energy modeling and leverages it to devise new efficient architectures and performing applications for sustainable future computing. We see this research as a contribution by computer science to reduce the energy hunger of compute resources in times of climate change. This modeling refers both to classical von Neumann architectures built into computing machines used in data centers and to embedded HPC architectures, e.g., required for autonomous driving. Furthermore, to consider computer architectures at the cutting edge, what is mandatory, too, accelerator cores for neuromorphic computing and new compute principles like near- and in-memory-computing are also in the focus. These architectures serve also as a platform for a spectrum of applications, which on the one hand cover a representative spectrum of computational patterns and on the other hand are typical for a certain architecture:
(i) AI and Machine Learning (ML), (ii) autonomous driving for embedded HPC, (iii) computer vision on near-memory architectures, and (iv) spiking neural network simulation to utilize classical computing with and without neuromorphic acceleration.
Holistic Energy and Performance Modeling for Sustainable Computing
A DFG Research Unit