|
ИСТИНА |
Войти в систему Регистрация |
ИСТИНА ЦЭМИ РАН |
||
Comparative analysis of oneAPI and CUDA technologies for the parallel implementation of matrix-free finite element methodтезисы доклада
- Авторы: Mokin Arseniy, Vershinin Anatoly, Sabinin Grigory, Levin Vladimir
- Сборник: Book of abstracts 15th World Congress on Computational Mechanics (WCCM-XV)
- Тезисы
- Год издания: 2022
- Издательство: International Center for Numerical Methods in Engineering
- Местоположение издательства: Spain
- Первая страница: 2056
- Последняя страница: 2056
- Аннотация: Modern problems [1] that use finite elements require substantial computing resources. One of the most efficient solutions in terms of the SIMD approach is the use of matrix-free methods on GPU [2]. This paper discusses the usage of the Element-by-element (EbE) method [3] for some 3D Solid Mechanics problems. The main idea of the EbE method is the absence of global stiffness matrix assembly. This results in high memory efficiency. The purpose of this work is to prove the possibility of a crossplatform implementation of the Finite Element Method using parallelism on high-performance hybrid architectures [4]. This is conducted by a comparative analysis of the performance of CUDA and oneApi technologies realizations. Obtained results make it possible to perform calculations faster than many adapted computational modern systems.
- Добавил в систему: Вершинин Анатолий Викторович