HIPERCONE FDTD: Vectorized highly scalable full-wave electromagnetic solver
Sergei Belousov, Sergey Khilkov, Vadim Levchenko, Anastasia Perepelkina, Ilya Valuev
Abstract
Sergei Belousov, Sergey Khilkov, Vadim Levchenko, Anastasia Perepelkina, Ilya Valuev
Abstract
In this paper we introduce a novel full-wave electromagnetic solver based on the Finite-Difference Time Domain method, which is extremely efficient in terms of CPU performance and scalability. These features of the HIPERCONE solver are attained by the use of asynchronous mesh updates, localization of data in the fast memory, and parallelism at all levels including vectorization. The algorithms in the solver are used to achieve the performance up to 1-2 orders of magnitude higher than the traditional approaches. Unlike the traditional memory-bound electromagnetic solvers, the maximal performance rate of HIPERCONE FDTD in terms of mesh cell updates per second is reached for large meshes occupying or even exceeding the total available CPU RAM. Therefore, the HIPERCONE solver is especially advantageous in solving large-scale problems. In this work we describe the algorithmic background of the simulation method and give an example of a typical large application which benefits from the solver's performance.
OpenAlex reports 1 citations for this work. Citation counts describe recorded attention and do not establish research quality.
A contribution statement is not available in the OpenAlex record.
Method details are not available in the OpenAlex metadata.
Findings are not separately available in the OpenAlex metadata.
Limitations are not available in the OpenAlex metadata.
Application details are not available in the OpenAlex metadata.
In this paper we introduce a novel full-wave electromagnetic solver based on the Finite-Difference Time Domain method, which is extremely efficient in terms of CPU performance and scalability. These features of the HIPERCONE solver are attained by the use of asynchronous mesh updates, localization of data in the fast memory, and parallelism at all levels including vectorization. The algorithms in the solver are used to achieve the performance up to 1-2 orders of magnitude higher than the traditional approaches. Unlike the traditional memory-bound electromagnetic solvers, the maximal performance rate of HIPERCONE FDTD in terms of mesh cell updates per second is reached for large meshes occupying or even exceeding the total available CPU RAM. Therefore, the HIPERCONE solver is especially advantageous in solving large-scale problems. In this work we describe the algorithmic background of the simulation method and give an example of a typical large application which benefits from the solver's performance.
Key concepts: Solver, Computer science, Parallel computing, Finite-difference time-domain method, Computational science, Scalability, Asynchronous communication, Polygon mesh