Interactive High Performance Volume Rendering

This thesis is about Direct Volume Rendering on high performance computing systems. As direct rendering methods do not create a lower-dimensional geometric representation, the whole scientific dataset must be kept in memory. Thus, this family of algorithms has a tremendous resource demand. Direct Volume Rendering algorithms in general are well suited to be implemented for dedicated graphics \nhardware. Nevertheless, high performance computing systems often do not provide resources for hardware accelerated rendering, so that the visualization algorithm must be implemented for the available general-purpose hardware. \n \nEver growing datasets that imply copying large amounts of data from the compute system to the workstation of the scientist, and the need to review intermediate simulation results, make porting Direct Volume Rendering to high performance computing systems highly relevant. The contribution of this thesis is twofold. \n \nAs part of the first contribution, after devising a software architecture for general implementations of Direct Volume Rendering on highly parallel platforms, parallelization issues and implementation details for various modern architectures are discussed. The contribution results in a highly parallel implementation that tackles several platforms. \n \nThe second contribution is concerned with the display phase of the “Distributed Volume Rendering Pipeline”. Rendering on a high performance computing system typically implies displaying the rendered result at a remote location. This thesis presents a remote rendering technique that is capable of hiding latency and can thus be used in an interactive environment.