Need Direction for multiGPU implementation

[Shweta Nayak]

Hi,
I have implemented a library for CG solver using CuSP which I am using in OpenSEES(earthquake simulator) for solving PDE. Now I want to convert it in multi GPU  format can you please give me the direction

[Steven Dalton]

Hello Shweta,

  This is a good question and you can approach multigpu support from a few different directions depending on your preferences and experience. The first way is to run cusp independently within each GPU and manually perform the inter-gpu copies using your own library or API calls. The second and possibly more involved but IMO cleaner way is to build your own custom backend for multigpus based on execution policies. The develop branch of Cusp is built around using execution policies as a flexible dispatching mechanism [1].

  The first approach has more examples and documentation online so it should be easier to get started and implement the functionality you require. The second approach is newer and will require more understanding of how Thrust/Cusp use execution policies internally.

Steve

[1] https://github.com/thrust/thrust/wiki/Derived-Systems

On Tue, Apr 19, 2016 at 3:00 AM, Shweta Nayak <nayaks...@gmail.com> wrote:

Hi,
I have implemented a library for CG solver using CuSP which I am using in OpenSEES(earthquake simulator) for solving PDE. Now I want to convert it in multi GPU  format can you please give me the direction

--
You received this message because you are subscribed to the Google Groups "cusp-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to cusp-users+...@googlegroups.com.
To post to this group, send email to cusp-...@googlegroups.com.
Visit this group at https://groups.google.com/group/cusp-users.
For more options, visit https://groups.google.com/d/optout.