Understanding how dynamic patterns vary across large spatiotemporal ensembles is essential in many scientific domains. In fluid dynamics, for instance, researchers analyze how splash patterns in droplet impact experiments change with physical parameters such as fluid type or impact velocity. These experiments produce large volumes of data where patterns differ in size, shape, and duration, making manual analysis tedious and error-prone. Recently, interactive visualization approaches have been developed to assist analysis using learned similarity models for pattern-based querying. However, they assume fixed-size inputs and only support single-pattern queries, thus limiting their effectiveness for multiscale, multi-pattern analysis and exploration of ensembles. In this paper, we present a visual analysis approach for the interactive exploration of spatiotemporal ensembles through multiscale pattern querying. Our approach extends an existing similarity model to support variable-sized patterns, allowing users to define queries by selecting examples directly on visualized data. Coordinated views enable interactive querying, comparison, and analysis of pattern occurrences and relate pattern occurrences to ensemble parameters. A guidance mechanism supports the user in finding underexplored regions. We demonstrate the utility of our approach on synthetic and real-world datasets. Domain expert feedback confirms that the approach is intuitive, easy to use, and effective for revealing parameter-pattern relationships.

Fluid simulation is a central topic in computer graphics, encompassing a wide range of methodologies for modeling Newtonian, non-Newtonian, and viscoelastic behaviors across both single-phase and multiphase settings. Existing single-phase frameworks have achieved high visual fidelity, yet multiphase simulations remain limited in accurately capturing complex phase interactions, particularly under high-viscosity-ratio or viscoelastic conditions. To address these challenges, we develop a unified multiphase viscoelastic formulation capable of handling diverse fluid types-including Newtonian, shear-dependent non-Newtonian, and viscoelastic flows-within a single consistent framework. The formulation extends mixture-model approaches through a multi-mode conformation tensor representation, which enhances numerical stability via phase-level stress corrections and efficiently captures a broad spectrum of rheological behaviors. Compared with existing techniques, our framework achieves improved momentum-mass consistency and numerical stability, maintaining physically plausible results across wide viscosity ranges, advancing the state of the art in multiphase viscoelastic fluid simulation.

Abstract Varying the input parameters of simulations or experiments often leads to different classes of results. Parameter sensitivity analysis in this context includes estimating the sensitivity to the individual parameters, that is, to understand which parameters contribute most to changes in output classifications and for which parameter ranges these occur. We propose a novel visual parameter sensitivity analysis approach based on Voronoi cell interfaces between the sample points in the parameter space to tackle the problem. The Voronoi diagram of the sample points in the parameter space is first calculated. We then extract Voronoi cell interfaces which we use to quantify the sensitivity to parameters, considering the class label information of each sample's corresponding output. Multiple visual encodings are then utilized to represent the cell interface transitions and class label distribution, including stacked graphs for local parameter sensitivity. We evaluate the approach's expressiveness and usefulness with case studies for synthetic and real‐world datasets.

Abstract Scatterplots are widely used in exploratory data analysis. Representing data points as glyphs is often crucial for in‐depth investigation, but this can lead to significant overlap and visual clutter. Recent post‐processing techniques address this issue, but their computational and/or visual scalability is generally limited to thousands of points and unable to effectively deal with large datasets in the order of millions. This paper introduces Sca2Gri (Scalable Gridified Scatterplots), a grid‐based post‐processing method designed for analysis scenarios where the number of data points substantially exceeds the number of glyphs that can be reasonably displayed. Sca2Gri enables interactive grid generation for large datasets, offering flexible user control of glyph size, maximum displacement for point to cell mapping, and scatterplot focus area. While Sca2Gri's computational complexity scales cubically with the number of cells (which is practically bound to thousands for legible glyph sizes), its complexity is linear with respect to the number of data points, making it highly scalable beyond millions of points.

Abstract We present HyperFLINT (Hypernetwork‐based FLow estimation and temporal INTerpolation), a novel deep learning‐based approach for estimating flow fields, temporally interpolating scalar fields, and facilitating parameter space exploration in spatio‐temporal scientific ensemble data. This work addresses the critical need to explicitly incorporate ensemble parameters into the learning process, as traditional methods often neglect these, limiting their ability to adapt to diverse simulation settings and provide meaningful insights into the data dynamics. HyperFLINT introduces a hypernetwork to account for simulation parameters, enabling it to generate accurate interpolations and flow fields for each timestep by dynamically adapting to varying conditions, thereby outperforming existing parameter‐agnostic approaches. The architecture features modular neural blocks with convolutional and deconvolutional layers, supported by a hypernetwork that generates weights for the main network, allowing the model to better capture intricate simulation dynamics. A series of experiments demonstrates HyperFLINT's significantly improved performance in flow field estimation and temporal interpolation, as well as its potential in enabling parameter space exploration, offering valuable insights into complex scientific ensembles.

We present FLINT (learning-based FLow estimation and temporal INTerpolation), a novel deep learning-based approach to estimate flow fields for 2D+time and 3D+time scientific ensemble data. FLINT can flexibly handle different types of scenarios with (1) a flow field being partially available for some members (e.g., omitted due to space constraints) or (2) no flow field being available at all (e.g., because it could not be acquired during an experiment). The design of our architecture allows to flexibly cater to both cases simply by adapting our modular loss functions, effectively treating the different scenarios as flow-supervised and flow-unsupervised problems, respectively (with respect to the presence or absence of ground-truth flow). To the best of our knowledge, FLINT is the first approach to perform flow estimation from scientific ensembles, generating a corresponding flow field for each discrete timestep, even in the absence of original flow information. Additionally, FLINT produces high-quality temporal interpolants between scalar fields. FLINT employs several neural blocks, each featuring several convolutional and deconvolutional layers. We demonstrate performance and accuracy for different usage scenarios with scientific ensembles from both simulations and experiments.

Bone cement filling is an important method for preventing osteoporosis and treating fractures. In bone cement filling surgery, the preparation and dosage of the cement usually depend on specific product manuals and the doctor’s experience. If bone cement is not used properly, it may cause additional damage. For teaching and auxiliary medical purposes, for example, assisting doctors to observe the possible flow of bone cement, this paper proposes a multiphase non-Newtonian fluid simulation method to simulate and visualize the flow behavior during the wet sand phase of bone cement blending and polymerization. Our method enables showing intuitively the application process of bone cement under different scene settings to obtain dynamic bone cement effects with high stability and performance. Compared with other methods, our method can simulate highly viscous mixed fluids efficiently and robustly, which supports our method’s usage in the aforementioned training and experimentation scenarios.

Abstract We propose an SPH‐based method for simulating viscoelastic non‐Newtonian fluids within a multiphase framework. For this, we use mixture models to handle component transport and conformation tensor methods to handle the fluid's viscoelastic stresses. In addition, we consider a bonding effects network to handle the impact of microscopic chemical bonds on phase transport. Our method supports the simulation of both steady‐state viscoelastic fluids and discontinuous shear behavior. Compared to previous work on single‐phase viscous non‐Newtonian fluids, our method can capture more complex behavior, including material mixing processes that generate non‐Newtonian fluids. We adopt a uniform set of variables to describe shear thinning, shear thickening, and ordinary Newtonian fluids while automatically calculating local rheology in inhomogeneous solutions. In addition, our method can simulate large viscosity ranges under explicit integration schemes, which typically requires implicit viscosity solvers under earlier single‐phase frameworks.

We developed a new approach comprised of different visualizations for the comparative spatio-temporal analysis of displacement processes in porous media. We aim to analyze and compare ensemble datasets from experiments to gain insight into the influence of different parameters on fluid flow. To capture the displacement of a defending fluid by an invading fluid, we first condense an input image series to a single time map. From this map, we generate a spatio-temporal flow graph covering the whole process. This graph is further simplified to only reflect topological changes in the movement of the invading fluid. Our interactive tools allow the visual analysis of these processes by visualizing the graph structure and the context of the experimental setup, as well as by providing charts for multiple metrics. We apply our approach to analyze and compare ensemble datasets jointly with domain experts, where we vary either fluid properties or the solid structure of the porous medium. We finally report the generated insights from the domain experts and discuss our contribution's advantages, generality, and limitations.

We present an efficient raycasting algorithm for rendering Volumetric Depth Images (VDIs), and we show how it can be used in a remote visualization setting with VDIs generated and streamed from a remote server. VDIs are compact view-dependent volume representations that enable interactive visualization of large volumes at high frame rates by decoupling viewpoint changes from expensive rendering calculations. However, current rendering approaches for VDIs struggle with achieving interactive frame rates at high image resolutions. Here, we exploit the properties of perspective projection to simplify intersections of rays with the view-dependent frustums in a VDI and leverage spatial smoothness in the volume data to minimize memory accesses. Benchmarks show that responsive frame rates can be achieved close to the viewpoint of generation for HD display resolutions, providing high-fidelity approximate renderings of Gigabyte-sized volumes. We also propose a method to subsample the VDI for preview rendering, maintaining high frame rates even for large viewpoint deviations. We provide our implementation as an extension of an established open-source visualization library.

Building on our original Virtual Ray Tracer tool, we present Virtual Ray Tracer 2.0, an interactive and gamified application that allows students/users to view and explore the ray tracing process in real-time. The application shows a scene containing a camera casting rays which interact with objects in the scene. Users are able to modify and explore ray properties such as their animation speed, the number of rays and their visual style, as well as the material properties of the objects in the scene. The goal of the application is to help the users – students of Computer Graphics and the general public – to better understand the ray tracing process and its characteristics. This includes not only the basics of ray tracing, but also more advanced concepts such as soft shadows. To invite users to learn and explore, various explanations and scenes are provided by the application at different levels of complexity, each with a step-by-step tutorial. Several user studies showed the effectiveness of the tool in supporting the understanding and teaching of ray tracing. The educational tool is built with the cross-platform engine Unity, and we make it fully available to be extended and/or adjusted to fit the requirements of courses at other institutions, educational tutorials, or of enthusiasts from the general public.

The visualization of results while the simulation is running is increasingly common in extreme scale computing environments. We present a novel approach for in situ generation of image databases to achieve cost savings on supercomputers. Our approach, a hybrid between traditional inline and in transit techniques, dynamically distributes visualization tasks between simulation nodes and visualization nodes, using probing as a basis to estimate rendering cost. Our hybrid design differs from previous works in that it creates opportunities to minimize idle time from four fundamental types of inefficiency: variability, limited scalability, overhead, and rightsizing. We demonstrate our results by comparing our method against both inline and in transit methods for a variety of configurations, including two simulation codes and a scaling study that goes above 19 K cores. Our findings show that our approach is superior in many configurations. As in situ visualization becomes increasingly ubiquitous, we believe our technique could lead to significant amounts of reclaimed cycles on supercomputers.

Abstract This paper introduces an optimization approach for generating grid layouts from large data collections such that they are amenable to level‐of‐detail presentation and exploration. Classic (flat) grid layouts visually do not scale to large collections, yielding overwhelming numbers of tiny member representations. The proposed local search‐based progressive optimization scheme generates hierarchical grids: leaves correspond to one grid cell and represent one member, while inner nodes cover a quadratic range of cells and convey an aggregate of contained members. The scheme is solely based on pairwise distances and jointly optimizes for homogeneity within inner nodes and across grid neighbors. The generated grids allow to present and flexibly explore the whole data collection with arbitrary local granularity. Diverse use cases featuring large data collections exemplify the application: stock market predictions from a Black‐Scholes model, channel structures in soil from Markov chain Monte Carlo, and image collections with feature vectors from neural network classification models. The paper presents feedback by a domain scientist, compares against previous approaches, and demonstrates visual and computational scalability to a million members, surpassing classic grid layout techniques by orders of magnitude.

Abstract We developed a new visualization approach to gain a better understanding of the displacement of one fluid phase by another in porous media. This is based on a recent experimental parameter study with varying capillary numbers and viscosity ratios. We analyse the temporal evolution of characteristic values in this two‐phase flow scenario and discuss how to directly compare experiments across different temporal scales. To enable spatio‐temporal analysis, we introduce a new abstract visual representation showing which paths through the porous medium were occupied and for how long. These transport networks allow to assess the impact of different acting forces and they are designed to yield expressive comparability and linking to the experimental parameter space both supported by additional visual cues. This joint work of porous media experts and visualization researchers yields new insights regarding two‐phase flow on the microscale, and our visualization approach contributes towards the overarching goal of the domain scientists to characterize porous media flow based on capillary numbers and viscosity ratios.

We introduce an ML-driven approach that enables interactive example-based queries for similar behavior in ensembles of spatiotemporal scientific data. This addresses an important use case in the visual exploration of simulation and experimental data, where data is often large, unlabeled and has no meaningful similarity measures available. We exploit the fact that nearby locations often exhibit similar behavior and train a Siamese Neural Network in a self-supervised fashion, learning an expressive latent space for spatiotemporal behavior. This space can be used to find similar behavior with just a few user-provided examples. We evaluate this approach on several ensemble datasets and compare with multiple existing methods, showing both qualitative and quantitative results.

Abstract We present our approach for the dense visualization and temporal exploration of moving and deforming shapes from scientific experiments and simulations. Our image space representation is created by convolving a noise texture along shape contours (akin to LIC). Beyond indicating spatial structure via luminosity, we additionally use colour to depict time or classes of shapes via automatically customized maps. This representation summarizes temporal evolution, and provides the basis for interactive user navigation in the spatial and temporal domain in combination with traditional renderings. Our efficient implementation supports the quick and progressive generation of our representation in parallel as well as adaptive temporal splits to reduce overlap. We discuss and demonstrate the utility of our approach using 2D and 3D scalar fields from experiments and simulations.

Collaborative exploration of scientific data sets across large high-resolution displays requires both high visual detail as well as low-latency transfer of image data (oftentimes inducing the need to trade one for the other). In this work, we present a system that dynamically adapts the encoding quality in such systems in a way that reduces the required bandwidth without impacting the details perceived by one or more observers. Humans perceive sharp, colourful details, in the small foveal region around the centre of the field of view, while information in the periphery is perceived blurred and colourless. We account for this by tracking the gaze of observers, and respectively adapting the quality parameter of each macroblock used by the H.264 encoder, considering the so-called visual acuity fall-off. This allows to substantially reduce the required bandwidth with barely noticeable changes in visual quality, which is crucial for collaborative analysis across display walls at different locations. We demonstrate the reduced overall required bandwidth and the high quality inside the foveated regions using particle rendering and parallel coordinates.

We present a machine learning-based approach for detecting and visualizing complex behavior in spatiotemporal volumes. For this, we train models to predict future data values at a given position based on the past values in its neighborhood, capturing common temporal behavior in the data. We then evaluate the model's prediction on the same data. High prediction error means that the local behavior was too complex, unique or uncertain to be accurately captured during training, indicating spatiotemporal regions with interesting behavior. By training several models of varying capacity, we are able to detect spatiotemporal regions of various complexities. We aggregate the obtained prediction errors into a time series or spatial volumes and visualize them together to highlight regions of unpredictable behavior and how they differ between the models. We demonstrate two further volumetric applications: adaptive timestep selection and analysis of ensemble dissimilarity. We apply our technique to datasets from multiple application domains and demonstrate that we are able to produce meaningful results while making minimal assumptions about the underlying data.

As our field matures, evaluation of visualization techniques has extended from reporting runtime performance to studying user behavior. Consequently, many methodologies and best practices for user studies have evolved. While maintaining interactivity continues to be crucial for the exploration of large data sets, no similar methodological foundation for evaluating runtime performance has been developed. Our analysis of 50 recent visualization papers on new or improved techniques for rendering volumes or particles indicates that only a very limited set of parameters like different data sets, camera paths, viewport sizes, and GPUs are investigated, which make comparison with other techniques or generalization to other parameter ranges at least questionable. To derive a deeper understanding of qualitative runtime behavior and quantitative parameter dependencies, we developed a framework for the most exhaustive performance evaluation of volume and particle visualization techniques that we are aware of, including millions of measurements on ten different GPUs. This paper reports on our insights from statistical analysis of this data, discussing independent and linear parameter behavior and non-obvious effects. We give recommendations for best practices when evaluating runtime performance of scientific visualization applications, which can serve as a starting point for more elaborate models of performance quantification.

We present a method for the spatio-temporal analysis of gaze data from multiple participants in the context of a video stimulus. For such data, an overview of the recorded patterns is important to identify common viewing behavior (such as attentional synchrony) and outliers. We adopt the approach of space-time cube visualization, which extends the spatial dimensions of the stimulus by time as the third dimension. Previous work mainly handled eye tracking data in the space-time cube as point cloud, providing no information about the stimulus context. This paper presents a novel visualization technique that combines gaze data, a dynamic stimulus, and optical flow with volume rendering to derive an overview of the data with contextual information. With specifically designed transfer functions, we emphasize different data aspects, making the visualization suitable for explorative analysis and for illustrative support of statistical findings alike.

We present a visualization approach for the analysis of CO2 bubble-induced attenuation in porous rock formations. As a basis for this, we introduce customized techniques to extract CO2 bubbles and their surrounding porous structure from X-ray computed tomography data (XCT) measurements. To understand how the structure of porous media influences the occurrence and the shape of formed bubbles, we automatically classify and relate them in terms of morphology and geometric features, and further directly support searching for promising porous structures. To allow for the meaningful direct visual comparison of bubbles and their structures, we propose a customized registration technique considering the bubble shape as well as its points of contact with the porous media surface. With our quantitative extraction of geometric bubble features, we further support the analysis as well as the creation of a physical model. We demonstrate that our approach was successfully used to answer several research questions in the domain, and discuss its high practical relevance to identify critical seismic characteristics of fluid-saturated rock that govern its capability to store CO2.

We present an approach for interactively analyzing large dynamic graphs consisting of several thousand time steps with a particular focus on temporal aspects. we employ a static representation of the time-varying graph based on the concept of space-time cubes, i.e., we create a volumetric representation of the graph by stacking the adjacency matrices of each of its time steps. To achieve an efficient analysis of complex data, we discuss three classes of analytics methods of particular importance in this context: data views, aggregation and filtering, and comparison. For these classes, we present a GPU-based implementation of respective analysis methods that enable the interactive analysis of large graphs. We demonstrate the utility as well as the scalability of our approach by presenting application examples for analyzing different time-varying data sets.

Abstract We visualize contours for spatio‐temporal processes to indicate where and when non‐continuous changes occur or spatial bounds are encountered. All time steps are comprised densely in one visualization, with contours allowing to efficiently analyze processes in the data even in case of spatial or temporal overlap. Contours are determined on the basis of deep raycasting that collects samples across time and depth along each ray. For each sample along a ray, its closest neighbors from adjacent rays are identified, considering time, depth, and value in the process. Large distances are represented as contours in image space, using color to indicate temporal occurrence. This contour representation can easily be combined with volume rendering‐based techniques, providing both full spatial detail for individual time steps and an outline of the whole time series in one view. Our view‐dependent technique supports efficient progressive computation, and requires no prior assumptions regarding the shape or nature of processes in the data. We discuss and demonstrate the performance and utility of our approach via a variety of data sets, comparison and combination with an alternative technique, and feedback by a domain scientist.

Abstract We present an approach to adaptively select time steps from time‐dependent volume data sets for an integrated and comprehensive visualization. This reduced set of time steps not only saves cost, but also allows to show both the spatial structure and temporal development in one combined rendering. Our selection optimizes the coverage of the complete data on the basis of a minimum‐cost flow‐based technique to determine meaningful distances between time steps. As both optimal solutions of the involved transport and selection problem are prohibitively expensive, we present new approaches that are significantly faster with only minor deviations. We further propose an adaptive scheme for the progressive incorporation of new time steps. An interactive volume raycaster produces an integrated rendering of the selected time steps, and their computed differences are visualized in a dedicated chart to provide additional temporal similarity information. We illustrate and discuss the utility of our approach by means of different data sets from measurements and simulation.

We present a novel technique to generate transformations between arbitrary volumes, providing both expressive distances and smooth interpolates. In contrast to conventional morphing or warping approaches, our technique requires no user guidance, intermediate representations (like extracted features), or blending, and imposes no restrictions regarding shape or structure. Our technique operates directly on the volumetric data representation, and while linear programming approaches could solve the underlying problem optimally, their polynomial complexity makes them infeasible for high-resolution volumes. We therefore propose a progressive refinement approach designed for parallel execution that is able to quickly deliver approximate results that are iteratively improved toward the optimum. On this basis, we further present a new approach for the streaming selection of time steps in temporal data that allows for the reconstruction of the full sequence with a user-specified error bound. We finally demonstrate the utility of our technique for different applications, compare our approach against alternatives, and evaluate its characteristics with a variety of different data sets.

We present an integrated approach for real-time performance prediction of volume raycasting that we employ for load balancing and sampling resolution tuning. In volume rendering, the usage of acceleration techniques such as empty space skipping and early ray termination, among others, can cause significant variations in rendering performance when users adjust the camera configuration or transfer function. These variations in rendering times may result in unpleasant effects such as jerky motions or abruptly reduced responsiveness during interactive exploration. To avoid those effects, we propose an integrated approach to adapt rendering parameters according to performance needs. We assess performance-relevant data on-the-fly, for which we propose a novel technique to estimate the impact of early ray termination. On the basis of this data, we introduce a hybrid model, to achieve accurate predictions with minimal computational footprint. Our hybrid model incorporates aspects from analytical performance modeling and machine learning, with the goal to combine their respective strengths. We show the applicability of our prediction model for two different use cases: (1) to dynamically steer the sampling density in object and/or image space and (2) to dynamically distribute the workload among several different parallel computing devices. Our approach allows to reliably meet performance requirements such as a user-defined frame rate, even in the case of sudden large changes to the transfer function or the camera orientation.

We present a novel scheme for progressive rendering in interactive visualization. Static settings with respect to a certain image quality or frame rate are inherently incapable of delivering both high frame rates for rapid changes and high image quality for detailed investigation. Our novel technique flexibly adapts by steering the visualization process in three major degrees of freedom: when to terminate the refinement of a frame in the background and start a new one, when to display a frame currently computed, and how much resources to consume. We base these decisions on the correlation of the errors due to insufficient sampling and response delay, which we estimate separately using fast yet expressive heuristics. To automate the configuration of the steering behavior, we employ offline video quality analysis. We provide an efficient implementation of our scheme for the application of volume raycasting, featuring integrated GPU-accelerated image reconstruction and error estimation. Our implementation performs an integral handling of the changes due to camera transforms, transfer function adaptations, as well as the progression of the data to in time. Finally, the overall technique is evaluated with an expert study.

The global scheduler of a current GPU distributes thread blocks to symmetric multiprocessors (SM), which schedule threads for execution with the granularity of a warp. Threads in a warp execute the same code path in lockstep, which potentially leads to a large amount of wasted cycles for divergent control flow. In order to overcome this general issue of SIMT architectures, we propose techniques to relax divergence on the fly within a computation kernel in order to achieve a much higher total utilization of processing cores. We propose techniques for branch and loop divergence (which may also be combined) switching to suitable tasks during a GPU kernel run every time divergence occurs. Our newly introduced techniques can easily be applied to arbitrary iterative algorithms and we evaluate the performance and effectiveness of our approach exemplarily via synthetic and real world applications.

This paper presents a visualization approach for detecting and exploring similarity in the temporal variation of field data. We provide an interactive technique for extracting correlations from similarity matrices which capture temporal similarity of univariate functions. We make use of the concept to extract periodic and quasiperiodic behavior at single (spatial) points as well as similarity between different locations within a field and also between different data sets. The obtained correlations are utilized for visual exploration of both temporal and spatial relationships in terms of temporal similarity. Our entire pipeline offers visual interaction and inspection, allowing for the flexibility that in particular time-dependent data analysis techniques require. We demonstrate the utility and versatility of our approach by applying our implementation to data from both simulation and measurement.

Molecular dynamics is a widely used simulation technique to investigate material properties and structural changes under external forces. The availability of more powerful clusters and algorithms continues to increase the spatial and temporal extents of the simulation domain. This poses a particular challenge for the visualization of the underlying processes which might consist of millions of particles and thousands of time steps. Some application domains have developed special visual metaphors to only represent the relevant information of such data sets but these approaches typically require detailed domain knowledge that might not always be available or applicable. We propose a general technique that replaces the huge amount of simulated particles by a smaller set of representatives that are used for the visualization instead. The representatives capture the characteristics of the underlying particle density and exhibit coherency over time. We introduce loose capacity-constrained Voronoi diagrams for the generation of these representatives by means of a GPU-friendly, parallel algorithm. This way we achieve visualizations that reflect the particle distribution and geometric structure of the original data very faithfully. We evaluate our approach using real-world data sets from the application domains of material science, thermodynamics and dynamical systems theory.

Abstract Higher‐order finite element methods have emerged as an important discretization scheme for simulation. They are increasingly used in contemporary numerical solvers, generating a new class of data that must be analyzed by scientists and engineers. Currently available visualization tools for this type of data are either batch oriented or limited to certain cell types and polynomial degrees. Other approaches approximate higher‐order data by resampling resulting in trade‐offs in interactivity and quality. To overcome these limitations, we have developed a distributed visualization system which allows for interactive exploration of non‐conforming unstructured grids, resulting from space‐time discontinuous Galerkin simulations, in which each cell has its own higher‐order polynomial solution. Our system employs GPU‐based raycasting for direct volume rendering of complex grids which feature non‐convex, curvilinear cells with varying polynomial degree. Frequency‐based adaptive sampling accounts for the high variations along rays. For distribution across a GPU cluster, the initial object‐space partitioning is determined by cell characteristics like the polynomial degree and is adapted at runtime by a load balancing mechanism. The performance and utility of our system is evaluated for different aeroacoustic simulations involving the propagation of shock fronts.

We present a development environment for distributed GPU computing targeted for multi-GPU systems, as well as graphics clusters. Our system is based on CUDA and logically extends its parallel programming model for graphics processors to higher levels of parallelism, namely, the PCI bus and network interconnects. While the extended API mimics the full function set of current graphics hardware-including the concept of global memory-on all distribution layers, the underlying communication mechanisms are handled transparently for the application developer. To allow for high scalability, in particular for network-interconnected environments, we introduce an automatic GPU-accelerated scheduling mechanism that is aware of data locality. This way, the overall amount of transmitted data can be heavily reduced, which leads to better GPU utilization and faster execution. We evaluate the performance and scalability of our system for bus and especially network-level parallelism on typical multi-GPU systems and graphics clusters.