clean-IT: Towards Sustainable Digital Technologiesclean-IT Initiative
Dieses Video gehört zum openHPI-Kurs clean-IT: Towards Sustainable Digital Technologies. Möchten Sie mehr sehen? Schreiben Sie sich kostenlos ein!

Lukas Wenzel (HPI) - Heterogeneous Computing

Zeitaufwand: etwa 25 Minuten
Sie verwenden unseren neuen Videoplayer. Sollten Sie Probleme haben, kontaktieren Sie bitte den Helpdesk. Sie können jederzeit zum alten Player wechseln.

Über dieses Video


As the share of renewable energy sources increases, computing systems and data centers will increasingly adapt their energy demand to the volatile production. At the HPI Operating Systems and Middleware Group, we research models and methods to understand and control a data center’s energy demand. In this endeavor, it is vital to understand how compute hardware consumes energy and how this behavior relates to different compute architectures. Mainstream compute hardware such as CPUs and GPUs offer a limited variety of microarchitectural tradeoffs, to accomodate different workload characteristics. However, this tradeoff is fixed, once a particular device is chosen and deployed. In contrast, Field Programmable Gate Arrays (FPGA) follow the fundamentally different approach of a reconfigurable microarchitecture. This enables a fine-grained adaption of the available hardware structures for a specific workload, offering a large potential for energy efficiency. More information...

Lukas Wenzel is a Ph.D. student and research assistant with Prof. Dr. Andreas Polze at the Operating Systems and Middleware Group. He received his B.Sc. and M.Sc. degrees from the Hasso-Plattner Institute, completing his master studies with a thesis in collaboration with IBM Research Böblingen on Operating System Facilities for FPGA Accelerator Designs. With a long-standing interest in compute hardware architecture, Lukas focuses on the domain of FPGA accelerators as a flexible platform to explore new architectural concepts for compute hardware with a view to reducing overheads and improving energy efficiency.