[mpich-discuss] How do you open (glut) windows on multiple MS Windows nodes for use with CUDA?

[Manuel Moser]

I'd like to use MPI for a network consisting of multiple nodes with Windows Vista. Each node has NVIDIA CUDA capable graphics card and a display attached to it. I'd like to compute, render and display graphics on all nodes by using the hardware on the respecting machines.

Is this possible with MPICH2? If yes, what do I have to do for it?

So far I've installed MPICH2 on all machines. I can sucessfully run command-line programs on multiple nodes. However if I start my own application with mpiexec.exe I can't open any window by using GLUT. glutMainLoop() starts and it is running but simply no window appears. Only if call mpiexec with the -localroot option, I can create a window, but of course only on the local machine and not on a remote machine.

I've noticed that usually when I query the OpenGL version I get something like the following if I run my application on a local machine: 
  GL_VERSION:   3.1.0
  GL_VENDOR:    NVIDIA Corporation
  GL_RENDERER:  GeForce 8800 GTX/PCI/SSE2

However if I'm using mpiexec I get the following:
  GL_VERSION:   1.1.0
  GL_VENDOR:    Microsoft Corporation
  GL_RENDERER:  GDI Generic

It seems when started with MPI my application doesn't have the right to open displays or access the graphics adapter.

Is there a way to configure MPI or Windows to allows me to execute programs which can open a window on all nodes with full access to the graphics adapter?

Or is there another solution which would allow me to do this, for example MS MPI? Or switching to Linux? Or should I use a completely different communication system instead?


Manuel
----
-- 
Jetzt kostenlos herunterladen: Internet Explorer 8 und Mozilla Firefox 3.5 -
sicherer, schneller und einfacher! http://portal.gmx.net/de/go/chbrowser