| Line 3: | Line 3: | ||
This paper aims to improve the e ciency of large-scale turbulent simulations by improving the arithmetic intensity of the operations. This is done by applying a parallel-in time ensemble averaging technique so that multiple ow states are run simultaneouslly in the same device. This transforms sparse matrix-vector products into sparse matrix-matrix products, improving the arithmetic intensity. The performance of these operations as well as the speed-ups generated in the operation itself, in the whole iteration and an estimation in the whole simulation is presented, so that for cases in which the averaging interval is signi cantlly longer than the transition interval, remarkable speed-ups in the whole iteration are obtained. | This paper aims to improve the e ciency of large-scale turbulent simulations by improving the arithmetic intensity of the operations. This is done by applying a parallel-in time ensemble averaging technique so that multiple ow states are run simultaneouslly in the same device. This transforms sparse matrix-vector products into sparse matrix-matrix products, improving the arithmetic intensity. The performance of these operations as well as the speed-ups generated in the operation itself, in the whole iteration and an estimation in the whole simulation is presented, so that for cases in which the averaging interval is signi cantlly longer than the transition interval, remarkable speed-ups in the whole iteration are obtained. | ||
| + | |||
| + | == Full Paper == | ||
| + | <pdf>Media:Draft_Sanchez Pinedo_344896000pap_2347.pdf</pdf> | ||
Revision as of 10:27, 23 October 2024
Abstract
This paper aims to improve the e ciency of large-scale turbulent simulations by improving the arithmetic intensity of the operations. This is done by applying a parallel-in time ensemble averaging technique so that multiple ow states are run simultaneouslly in the same device. This transforms sparse matrix-vector products into sparse matrix-matrix products, improving the arithmetic intensity. The performance of these operations as well as the speed-ups generated in the operation itself, in the whole iteration and an estimation in the whole simulation is presented, so that for cases in which the averaging interval is signi cantlly longer than the transition interval, remarkable speed-ups in the whole iteration are obtained.
Full Paper
Document information
Published on 23/10/24
Submitted on 23/10/24
Volume Advanced Parallel Algorithms for Extreme-Scale Simulations, 2024
DOI: 10.23967/eccomas.2024.015
Licence: CC BY-NC-SA license
Share this document
Keywords
claim authorship
Are you one of the authors of this document?