A parallel isosurface extraction component for visualization pipelines executing on GPU clusters

Isosurface extraction is a fundamental operation for the analysis and visualization of 3D scalar fields, and the marching cubes technique is the de facto standard to implement it. The processing of large datasets can be a very time-consuming task, and for this reason a number of parallel implementations have been proposed in the literature. In this paper we present a parallel implementation of the marching cubes algorithm that is able to exploit multiple CUDA devices without the need of any preprocessing operation for the dataset partitioning. The output format is suitable for direct visualization as well as for efficient storage; therefore it can be effectively used as a software component in visualization pipelines. Results obtained on a real GPU cluster show speedups of over 230 for the full algorithm (including file I/O and initializations) and of close to 1490 when considering only actual computations and data movements between hosts and devices. © 2014 Elsevier B.V. All rights reserved.