Visualization on JURECA

JURECA Visualization Nodes

JURECA contains two login visualization nodes and a partition of ten visualization nodes integrated in the batch system. All vis nodes are dedicated to GPU-assisted remote data visualization. Ten visualization nodes with 512 GB main memory and two nodes with 1024 GB main memory for memory demanding visualization or other pre- or post-processing tasks are offered. Each vis node is equipped with two NVIDIA Tesla K40 GPUs.

Here is an overview about the configuration:

6 visualization login nodes:

  • ssh access via jurecavis.fz-juelich.de (jurecavis[01-06].fz-juelich.de in round-robin fashion)
  • 512 GB main memory, 2 NVIDIA Tesla K40 GPUs
  • login nodes, that means direct access via ssh possible
  • shared by all users logged in

4 visualization batch nodes with 512 GB main memory:

  • 512 GB main memory, 2 NVIDIA Tesla K40 GPUs
  • access via batch system, partition vis
  • X server needs to be started for remote rendering with option --start-xserver
  • typical arguments for sbatch or salloc: -p vis --gres=mem512,gpu:2 --start-xserver
  • at most 4 nodes can be allocated in one batch job

2 visualization batch nodes with 1024 GB main memory:

  • 1024 GB main memory, 2 NVIDIA Tesla K40 GPUs
  • access via batch system, partition vis
  • X server needs to be started for remote rendering with option --start-xserver
  • typical arguments for sbatch or salloc: -p vis --gres=mem1024,gpu:2 --start-xserver

How to Use the Visualization Nodes

A graphical login simplifies the access to remote visualization and the data stored on the file system. Hence, we support and recommend the use of Virtual Network Computing (VNC) on the visualization nodes of JURECA. VNC provides a complete graphical and remote desktop by

  • transmitting the keyboard and mouse events from your computer to the remote machine and
  • relaying the graphical screen updates back in the other direction.

VNC is

  • platform-independent (VNC viewers are available for almost any operating system)
  • application-independent (no need for remote visualization support inside the application)

With the help of VirtualGL installed on JURECA vis nodes OpenGL commands of visualization software can be processed on their local GPUs. In contrast to X-forwarding (ssh -X) this can speed up visualization tremendously especially for large data sets.

For further details about how to use the vis nodes of JURECA for remote rendering please refer to the visualization related documentation at https://trac.version.fz-juelich.de/vis/

Additional Documentation

Remote 3D Visualization at JSC (PDF)