Hi, Bruno, Thank you for your insightful and significant reply with the PR link.
Assembly by GPU is still undergoing! This certainly is my most wished-for feature.😀 If currently the GPU is used for linear solver, then I think Kokkos leverage will be more natural for Trilinos ML library and Trilinos Vectors. Hope it isn't too much harder than compiling Trilinos with Kokkos. Tim, Sincerely On Mon, Oct 2, 2023 at 4:37 PM Bruno Turcksin <[email protected]> wrote: > Tim, > > There is currently no way to do what you want. There is ongoing work to > replace the Epetra wrappers with Tpetra ( > https://github.com/dealii/dealii/pull/16052), this will allow to use GPU > in matrix-based code. You will probably get the majority of the speedup > using GPU for the solver, so you could assemble the matrix on the CPU, move > it to GPU, and solve the system on the GPU. Unfortunately for now, you have > to do that yourself. > > Best, > > Bruno > > On Monday, October 2, 2023 at 4:09:01 AM UTC-4 Tim hyvärinen wrote: > >> Hi, deal community and developers, >> >> Thank you for all the help I ever got from you guys. >> >> I'm developing a MPI vector-valued non-linear PDEs solver. The way I >> chose is dumping newton iteration, which uses AMG preconditioner support >> from Trilinos 12.18 for linear solver. Currently, the program is CPU code. >> It works, and fast enough for practical purposes. >> >> However, as the subject suggests, I always wonder if I can take advantage >> of GPU to make my code faster or cheaper. The clusters I am using have >> Nvidia Ampere A100 and AMD MI215 gpu nodes. It will be a big waste if I >> don't use them. >> >> I know dealii has CUDA wrapper and Kokkos leverage. But what I don't know >> is how I can *properly* use them to speed up matrix-based newton >> iteration code. I wonder if there is a *suggested pattern* to conduct GPU >> programming in dealII? A naive idea in my mind is to use GPU for >> system-matrix assembly, but I didn't see this in the only cuda tutorial >> i.e., step-64. As there is possibly(always) an untold conversion for a >> legendary library. I wonder what are the suggested ways in dea.ll for using >> GPU in matrix-based code? >> >> Tim, >> Sincerely >> > -- > 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 [email protected]. > To view this discussion on the web visit > https://groups.google.com/d/msgid/dealii/c22dd877-6db1-416e-b6a4-c045242c3fddn%40googlegroups.com > <https://groups.google.com/d/msgid/dealii/c22dd877-6db1-416e-b6a4-c045242c3fddn%40googlegroups.com?utm_medium=email&utm_source=footer> > . > -- 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 [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/dealii/CAArwj0Fk%3DrGaSgJiAc2RPaMzxD%2ByA1Z%3D52C1YPmPF7vo5q8kMg%40mail.gmail.com.
