Launching pvserver with Nvidia Index and selecting the remote GPU to run on

Situation

I am setting up a reverse client-server connection connection using the Nvidia Index plugin. I use the plugin version 2.3 required for ParaView 5.7.0 according to the user’s guide taken from the Paraview download area (pdf).
The server installation has been compiled with EGL, so purely as a server machine.


Interpretation issue

Now, I have some difficulties in interpreting this line of the user’s guide using the information from pvserver --help:

mpirun -bynode -np 1 pvserver -display :0.0 --use-offscreen-rendering

I will go from the simplest to the most difficult

  • -bynode of mpi has been superseded by --map-by node; sorted
  • --use-off-screen of paraview has been supersed by --force-off-screen; sorted
  • what is then -display :0.0?

There is no similar option explained in the pvserver --help.

  1. What is the syntax of that :0.0?
  2. When can I get a description about this?

Action issues

Onto the practicalities: I do manage to open a working connection between client and server. At the server side I have 3 GPUs: one lightweight and two heavyweights. I know that with my licence I can only activate one.

The system starts the process on the weakest device, though. In contrast, I would like to have one of the more powerful GPUs to work for me.

  1. How do I tell the plugin-pvserver pair to use the GPU number 1 rather than number 0, for example? The numbering I am using here is the one of nvidia-smi, the management interface.

Thanks for explaining this

What is the syntax of that :0.0 ?

It is related to the DISPLAY variable a Xorg env var, which is different when using egl.

With EGL, the syntax should be pvserver --egl-device-index=1

@danlipsa @lhofmann