Distributed Parallelization of xPU Stencil Computations in Julia
We present a straightforward approach for distributed parallelization of stencil-based xPU applications on a regular staggered grid, which is instantiated in the package ImplicitGlobalGrid.jl. The approach allows to leverage remote direct memory access and enables close to ideal weak scaling of real-world applications on thousands of GPUs. The communication costs can be easily hidden behind computation.
Authors
- Ludovic Räss (ORCID: https://orcid.org/0000-0002-1136-899X)
- Samuel Omlin (ORCID: https://orcid.org/0000-0001-8969-5619)
- Ivan Utkin (ORCID: https://orcid.org/0000-0002-5233-6783)
Institutions
- Swiss Federal Institute for Forest, Snow and Landscape Research (CH)
- ETH Zurich (CH)
- Swisscom (Switzerland) (CH)
Publication Details
- Journal
- JuliaCon Proceedings
- Published
- 2024-11-21
- DOI
- https://doi.org/10.21105/jcon.00137
- Citations
- 2
- Primary Topic
- Parallel Computing and Optimization Techniques
- Type
- article
- Field-Weighted Citation Impact
- 0.98
Funders
- Centro Svizzero di Calcolo Scientifico
- National Supercomputing Centre Singapore