CPU parallel on multi-computers

I noted DualSPHysics is based on OpenMP, can this tool also support CPU parallel calculations on multi-computers?

Comments

  • I cannot exactly answer this question but I can write down a reply I got when I asked about MPI parallelization capabilities:
    Before going to MPI, you have to be sure that you really need that...
    First: OpenMP allows you to use 8-16 cores of a good CPU as the current ones...
    Second: You can use the GPU version of our code.. that will be always more efficicent and chepar than machines with MPI.... you can buy a good GPU card for 500-600 euros...

    We have only worked with MPI in the past in order to communicate different GPU cards hosted by different CPU that need MPI...
    Based on this, 1. MPI is not implemented. MPI is often used for implementing parallelization on computational clusters. 2. OpenMP is only intended to communicate between different cores on a singular CPU. 3. GPU performs considerably better than a CPU (For me a factor 6 faster).

    I hope this helps somewhat otherwise you will have to wait for someone more experienced to answer.
  • @bladomas1 Many thanks for the kind advice. I search on line and found a presentation on this topic. It seems there is some thing on this topic.

    http://www.archer.ac.uk/training/virtual/2018-01-24/Presentation_ecse_pdf.pdf

    In the presentation, It mentioned the reasons why MPI is necessary.
  • but in that work, UMAN team had access to a large cluster machine where they can use MPI...

    in your case, what are you going to use as a execution device? a PC or a cluster?
  • @Alex My university has a cluster that I can use. I heard from some colleague that the CPU core is more powerful than GPU core. Is that possible is the MPI implementstion can be shared so I can do comparations between CPU cluster and GPU?
Sign In or Register to comment.