Energy-Efficient Heterogenous Computing with SYNERGY

Energy-efficient computing aims to maximize the energy efficiency of computing systems. A key approach in modern parallel hardware is frequency scaling, enabled by Dynamic Voltage and Frequency Scaling. AMD, Intel, and NVIDIA GPUs provide APIs for power management and core frequency adjustment. Leveraging these capabilities allows optimizing both performance and energy consumption in GPU programs. In this presentation, Biagio will introduce SYnergy, a novel energy-efficient approach based on SYCL that leverages on frequency scaling to achieve unprecedented fine-grained energy savings on large-scale heterogeneous clusters. SYnergy was recently accepted for publication at SuperComputing 2023 and has been installed at CINECA Supercomputing.

Download Presentation


Learn about joining the UXL Foundation:

Join now