Efficient high-fidelity simulations for the energy transition using ARM and x86 64 architectures
Please login to view abstract download link
High-performance scientific computing is at the heart of digital technology. It enables the simulation of complex physical phenomena in a reasonable time. The current trend for computer hardware architecture is to shift towards more parallelism. To use these systems in the most optimal way different levels of parallelism need to be studied. The demand for a fast solution of simulations coupled with new computing architectures drives the need for an adequate programming model, challenging parallel algorithms, a suitable memory layout, appropriate architecture-specific parameters and a flexible software architecture. This paper focuses on two prominent x86 and ARM hardware architectures. First, it highlights the key considerations for achieving more efficient computation at the node level. It also evaluates whether optimal use of the mentioned computing resources can be possible by considering all levels of parallelism and how. It addresses the aforementioned aspects, through an iterative preconditioned linear solver for sparse, non-symmetric matrices. The scalability and time-to-solution of several test cases are demonstrated and analyzed.
