The strategies of promoting HPC efficiency

HPC applications have always been the main force in industrial simulations, and their performance directly affects the development progress of products and projects. However, even today’s huge scale of computing power cannot fully meet some HPC applications; and still need to sacrifice accuracy or fidelity to satisfy the industrial design. So, making HPC applications more efficient with the current computing power has become a major challenge. This session will discuss how to deal with the above challenges from two directions. The first is to accelerate the application performance through various optimization methods. The VASP (material science) will be taken as an example to introduce how to improve efficiency by optimizing parallel algorithms, hotspot functions, and IO processes. The second is the integration of AI algorithms and traditional HPC applications, such as weather prediction, material simulation, and computational fluid mechanics (CFD). We will present our research work in detail with CFD applications.