Daniel,
Using MPI and CUDA together is pretty much work in progress. On the > developer branch, LinearAlgebra::distributed::Vector supports > storing its elements on a CUDA device, but the values are copied to the > CPU for MPI communication. > The pull request https://github.com/dealii/dealii/pull/7303 allows to use > CUDA-aware MPI instead but the memory transfer needs to be optimized. > If I could get involved in this development, I would be helpful to optimize the memory transfer between GPU and CPU by testing different CUDA C programming schemes like this: https://devblogs.nvidia.com/how-overlap-data-transfers-cuda-cc/ In fact, as a principal investigator at Taiwan, I am qualified to submit jobs into Taiwania, a recently established super computing machine established in 2017, equipped with NVIDIA Tesla P100 SXM2. Moreover, sometime next year, we will have another super computing machine available to use (Taiwania 2), which now ranks 20th in the world with NVIDIA Tesla V100 SXM2. https://www.top500.org/system/179590 Sincerely, Chih-Che -- The deal.II project is located at http://www.dealii.org/ For mailing list/forum options, see https://groups.google.com/d/forum/dealii?hl=en --- You received this message because you are subscribed to the Google Groups "deal.II User Group" group. To unsubscribe from this group and stop receiving emails from it, send an email to dealii+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
