Energy-efficient multigrid smoothers and grid transfer operators on multi-core and GPU clusters

• Verfasst von: Wlotzka, Martin [VerfasserIn]
• Verfasst von: Heuveline, Vincent [VerfasserIn]
• Titel: Energy-efficient multigrid smoothers and grid transfer operators on multi-core and GPU clusters
• Verf.angabe: Martin Wlotzka, Vincent Heuveline
• Jahr: 2017
• Jahr des Originals: 2016
• Umfang: 12 S.
• Teil: volume:100
• Teil: year:2017
• Teil: pages:181-192
• Teil: extent:12
• Fussnoten: Gesehen am 01.10.2018
• Titel Quelle: Enthalten in: Journal of parallel and distributed computing
• Ort Quelle: Amsterdam [u.a.] : Elsevier, 1984
• Jahr Quelle: 2017
• Band/Heft Quelle: 100(2017), Seite 181-192
• DOI: doi:10.1016/j.jpdc.2016.05.006
• Datenträger: Online-Ressource
• Sprache: eng
• Sach-SW: Energy-aware numerics
• Sach-SW: Geometric multigrid
• Sach-SW: Heterogeneous platforms
• Sach-SW: Hybrid parallelization
• Sach-SW: Performance and energy assessment
• K10plus-PPN: 1581442211

Abstract

We investigate time and energy to solution for the CPU- and GPU-based execution of the compute intensive smoother and grid transfer operators in a geometric multigrid linear solver. We use a hybrid parallel implementation for both shared and distributed memory multi-core host systems comprising CUDA-capable devices. Our numerical experiments are designed to assess the effect of combining an MPI-parallel multigrid framework with OpenMP host threads or CUDA accelerators instead of MPI-only CPU computations for various parallel setups. We present runtime and energy measurements from a quad-CPU test system equipped with two GPUs. We find that using an accelerated asynchronous smoother can yield substantial savings of time and energy to solution over using a host-only Jacobi smoother in small and medium sized host systems with one or two multi-core CPUs. The acceleration of the grid transfer operators also yields a benefit, yet smaller than the benefit from the smoother. For large host systems a hybrid MPI-OpenMP parallelization turns out to be most beneficial with respect to energy consumption, although it is not the fastest option.