Session

DescriptionIn scientific computing and beyond, we are all used to optimizing time-to-solution in our computations and to use computers to their fullest power. Yet, the world is facing a climate crisis and we must all strive to minimize our energy footprint. The power consumption constraint for large-scale computing (aka exascale computing limited by 20MW) encourages scientists to revise the architectural design of hardware but also of applications, their underlying algorithms, as well as the working/ storage precision. By improving energy efficiency, we contribute to sustainable scientific computing. Developing energy-efficient scientific computing applications is nontrivial and requires expertise in several different areas: algorithms and applications, programming languages and compilers, numerical verification, and computer architecture. In this minisymposium, we aim to discuss challenges in designing and implementing energy-efficient mixed-precision algorithms with the assistance of computer arithmetic tools, energy efficiency modeling, hardware-software co-design in light of the end of Moore's law, and potential for automating energy-saving techniques. Key topics include establishing energy-to-solution as a HPC metric following our initial attempt https://zenodo.org/records/13306639, exploring energy optimization opportunities from algorithmic derivation and across computing stacks, and enhancing tools to aid energy-efficient software development.