[Paraview] //Remote rendering with tesla cards

[Benjamin Ahrenholz]

Dear all,

We are trying to utilize our cluster in order to visualize data directly on
the compute nodes. 24 nodes (each has 8 computing cores) are equipped with 4
Tesla cards. We were following the instructions given by the ParaView wiki
and by those slides here:

www.vpac.org/files/GPU-Slides/Visualisation_using_the_GPU_Computing_Processo
r.pdf

Meaning that Xservers are installed and glxgears is giving good results on
the remote nodes. However, if we try to use ParaView on the headnode
(assuming pvserver is running on the remote nodes, just using one GPU per
node:

#PBS -l nodes=dnode161:ppn=2+dnode162:ppn=2

/hpc3lustre/software/lib/openmpi-1.4.1-gcc41/bin/mpirun -np 2 \
/hpc3lustre/software/irmb/paraview/ParaView-3.9.0-dev/bin/pvserver \
-display :0.0 --use-offscreen-rendering

), we were confronted with the following errors after connecting from the
client to the servers:

Waiting for server...
X Error: GLXBadContextTag 173
  Extension:    157 (Uknown extension)
  Minor opcode: 5 (Unknown request)
  Resource id:  0x40010b0

...

X Error: GLXBadContextTag 173
  Extension:    157 (Uknown extension)
  Minor opcode: 145 (Unknown request)
  Resource id:  0x1
Xlib: sequence lost (0x10000 > 0x693c) in reply type 0x1!
Write failed: Broken pipe

Probably we are just missing a tiny thing, but looking for solution yet
yielded no results :( Maybe one of you had the same problems and some more
exprience? Also I think, the problem is not located at the remote nodes but
on the headnode. I am looking forward for any hints and suggestions. Thanks
in advance!

Ben.

P.S.: When does ParaView will get a "reload"-pipeline button?