Data Locality Optimization Strategies for AMR Applications on GPU - accelerated

Abstract

As the memory hierarchies of supercomputers get more complex, improving the performance of applications bounded by the data movement throughput becomes challenging. For example, TokyoTech’s TSUBAME, the machine we intend to use, includes different data transfer bottlenecks that complicate the domain decomposition and load balancing: inter-node connection… (More)

Topics

3 Figures and Tables

Cite this paper

@inproceedings{Wahib2017DataLO, title={Data Locality Optimization Strategies for AMR Applications on GPU - accelerated}, author={Mohamed Wahib}, year={2017} }