Powercapping the DTC: Conclusions and Lessons Learned¶
Johannes Gebert (HLRS)
Abstract¶
As the pace of Moore’s Law diminishes, high-performance computing (HPC) centers face the dual challenges of escalating energy consumption and the necessity for sustainable operations. To address these issues, innovative techniques are crucial for reducing energy usage while maintaining, or even enhancing, computational performance. DEPO—a dynamic, node-level power-capping approach independent of specific software, developed by Krzywaniak, Czarnul, and Proficz (2022)—offers a promising solution. In our study, we apply DEPO to real-world scenarios, particularly within the context of Finite Element Analysis (FEA). By employing Direct Tensor Computation (DTC) as developed by Ralf Schneider and Johannes Gebert, we explore DEPO’s practical applicability. We are also investigating the conditions under which powercapping should be used and the future prospects for this research.