Research Topic
Improving Energy Efficiency in Scientific Workloads on AI Centered HPC Hardware through Reduced Precision.
Supervisors
- Robert Schoene, PhD (Dresden University of Technology)
- Professor Jeronimo Castrillon (Dresden University of Technology)
- Professor Wolfgang Nagel (Dresden University of Technology)
Thesis Abstract
This PhD thesis explores how reduced and mixed precision can be used to make scientific applications more energy efficient on modern HPC systems. The work focuses on understanding when lower precision is actually useful, how it affects accuracy and performance, and how well current AI-oriented hardware supports these approaches. The goal is to identify practical ways of reducing energy consumption without unnecessarily sacrificing the quality of scientific results.
Publications
- Gülçin Gedik, Robert Schoene, and Roman Iakymchuk. “Mixed-Precision For Energy Efficient Computations”. In: SC25@WHPC Proceedings (Jan. 2025).
- Kajol Kulkarni et al. “Harvesting energy consumption on European HPC systems: Sharing Experience from the CEEC project”. In: HPCA EE HPC WG State of Practice Workshop (Jan. 2026).
Conference Presentations
- ISC Poster Presentation 2025
- NHR Conference Poster Presentations in 2025 & 2026
- ICPE Conference Paper Presentation
- NHR Conference Invited Talk 2026