Nikola Schuster (Stellenbosch University, Afrique du Sud) donnera l’exposé suivant

Title:
Energy-Efficient Processing in Radio Astronomy: Comparative Analysis
of Computing platforms and their Implementation

Short abstract:
As the Square Kilometer Array pushes the limits of radio astronomy,
energy-efficient computing is crucial. My research firstly compares
the energy efficiency of Field-Programmable Gate Arrays (FPGAs) and
Graphics Processing Units (GPUs), two key technologies in this field,
and secondly investigates the effect of using the HDL coder
abstraction layer.

First, we ran a correlator algorithm on both platforms and found that
FPGAs processed 127 times more instructions per kilowatt-hour than
GPUs, indicating significantly better energy efficiency. Secondly, we
examined the impact of using HDL Coder  an abstraction tool for
programming FPGAs. We used the correlator algorithm, and found that
the abstraction layer increased power consumption by 11.18%.

These results suggest that while GPUs are being adopted, FPGAs may
still offer superior energy efficiency for certain tasks. Moreover,
abstraction tools, though useful, come with an energy cost.