cuda+mpi

Some people use cuda+mpi ?
How could you build your code with cuda+mpi !?
Ex: mpi.cpp + cuda.cu ? or only in .cu !?

My os is Windows HPC Server 2008 ,I just want your views.

My Hardware is Tesla C1060 +Quadro NVS 295
I run example by nvidia GPU Computing SDK (Device Query)

When only run Device Query , I get some message…

There are 2 devices supporting CUDA …
Device 0 : "Tesla C1060 "…
Device 1 : “Quadro NVS 295”…

But when I mpi+cuda , I get message…

Hi, I am CPU 0
There is no device supporting CUDA.
Device 0: “Device Emulation (CPU)”

What are you views ?

I already find the answer…

Please read the paper…

http://resourcekit.windowshpc.net/MORE_INF…ver%202008.html

maybe this is the right link: http://resourcekit.windowshpc.net/MORE_INFO/33b-GPU_Computing_in_Windows_HPC_Server_2008.html

this may help…

http://forums.nvidia.com/lofiversion/index.php?t96673.html

in order for this to work, you need to modify the make file to substitute mpicc for gcc and g++. This provides the appropriate complier linking. From there you can can simply write your mpi code and add the command line arg listed in the link.

Let me know if that helps or if there are specific details that I left out.