MPI, UPC/Coarray Fortran on Phi

MPI, UPC/Coarray Fortran on Phi

I would like to compile and run software that uses MPI directly on a Phi card; that is, in native or symmetric mode.  I have a license for Intel Composer XE 2013.

What additional software do I have to purchase, if any, in order to run MPI codes on a single Phi card?  On two Phi cards in one host?  On two Phi cards installed in separate hosts (i.e. on a distributed memory cluster)?

I am also interested in running Unified Parallel C (UPC) and Coarray Fortran codes in these configurations.  Are these programming models supported on Phi?  If so, are there any limitations that I should be aware of?

4 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

The MPI implementations don't require any additonal hardware for running 1 or more coprocessor cards on a single host, although it's said that infiniband might improve performance of symmetric mode.

I haven't heard about results of the Intel Fortran co-array implementation on coprocessor.

SGI announced an implementation of UPC, but I haven't seen any results.

Host Windows implementation won't support MPI symmetric mode.

Thanks for your answers regarding UPC and Coarray Fortran.  It would be interesting to find out more about Coarray Fortran on Phi, since recent versions of the Intel Fortran compiler support it and the implementation (both for shared and distributed memory systems) uses Intel MPI behind the scenes.

The MPI part of my question boils down to this: Do I have to buy Intel MPI in order to run a single MPI job across two Phi cards running in native mode that are installed in two Linux hosts which are connected over a netwok?

Intel MPI is no longer being marketed separately. It is bundled in the Intel Cluster XE product.

Leave a Comment

Please sign in to add a comment. Not a member? Join today