ParaView and Offscreen Rendering

ParaView is often used to render visualization results. While in most cases, the results are presented to the users on their screens or monitors, there are cases when this is not required. For example, if running a batch script to generate several images, one may not care to see windows flashing in front of them while the results are being generated. Another use-case is when using a pvserver or pvrenderserver to render images remotely and having them sent back to the client for viewing. The server-side windows are do not serve any useful purpose in that mode. In other cases, showing a window may not be possible at all, for example on certain Linux clusters, one may not have access to an X server at all, in which case on-screen rendering is just not possible (this is also referred to as headless operating mode).

This page documents the onscreen, offscreen, and headless rendering support in ParaView.

Caveats

In this page, we limit ourselves to a Linux-based system. On Windows and macOS, offscreen support is available, but true headless operation is currently not widely used and hence left out of this discussion.

Terminology

A brief explanation of the terms:

Onscreen and offscreen support is built by default. Thus ParaView binaries available from paraview.org support these modes. Headless support requires special builds of ParaView which are not compatible with standard build options. Both version are available as a binary from our downloads page.

OpenGL Implementations

ParaView uses OpenGL for rendering. OpenGL is an API specification for 2D/3D rendering. Many vendors provide implementations of this API, including those that build GPUs.

For sake of simplicity, let's classify OpenGL implementations as hardware (H/W) and software (S/W). H/W includes OpenGL implementations provided by NVIDIA, ATI, Intel, Apple and others which typically use the system hardware infrastructure for rendering. The runtime libraries needed for these are available on the system. S/W currently includes Mesa3D – a software implementation of the OpenGL standard. Despite the names, H/W doesn't necessarily imply use of GPUs, nor does S/W imply exclusion of GPUs. Nonetheless we use this naming scheme as it has been prevalent.

APIs for Headless Support

Traditionally, OpenGL implementations are coupled with the window system to provide an OpenGL context. Thus, they are designed for non-headless operation. When it comes to headless operation, there are alternative APIs that an application can use to create the OpenGL context that avoid this dependency on the window system (or X server for the sake of this discussion).

Currently, ParaView supports two distinct APIs that are available for headless operation: EGL and OSMesa (also called Offscreen Mesa). It must be noted that headless support is a rapidly evolving area and changes are expected in coming months. Modern H/W OpenGL implementations support EGL while S/W (or Mesa) supports OSMesa. One has to build ParaView with specific CMake flags changed to enable either of these APIs. Which headless API you choose in your build depends on which OpenGL implementation you plan to use.

ParaView Builds

Before we look at the various ways you can build and use ParaView, let's summarize relevant CMake options available:

VTK_USE_X and VTK_OPENGL_HAS_OSMESA are mutually exclusive i.e. only one of the two can be ON at the same time. This is due to the fact that OSMesa does not support on-screen rendering and VTK's OpenGL selection is at build time.

VTK_OPENGL_HAS_EGL and VTK_OPENGL_HAS_OSMESA are mutually exclusive i.e. only one of the two can be ON at the same time.

A few things to note:

Default Rendering Modes

Depending of the enabled build options, which type of render window the ParaView executable creates by default also needs some explanation.

--use-offscreen-rendering command line option supported by ParaView 5.4 and earlier has now been deprecated and is interpreted as --force-offscreen-rendering.