Ray casting on shared-memory architectures: memory-hierarchy considerations in volume rendering

Palmer, Michael E.; Totty, Brian; Taylor, Stephen

doi:10.1109/4434.656777

Cited by 8 publications

(7 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Output Resolution Low High Small (Cabral et al, 1994) (Grzeszczuk et al, 1998) (Ikits et al, 2004) (Kruger and Westermann, 2003) ( Lacroute and Levoy, 1994) (Levoy, 1988) (Pfister et al, 1999) (Rezk-Salama et al, 2000 ( Westover, 1990) (McCorquodale andLombeyda, 2003) (Schwarz et al, 2004) Large (Bajaj et al, 2000) (Camahort and Chakravarty, 1993) (Elvins, 1992) (Hsu, 1993) (Ino et al, 2003) (Lombeyda et al, 2001) (Palmer et al, 1998) (Peterka et al, 2008) (Ma et al, 1994) (Marchesin et al, 2008) (Mller et al, 2006 (Muraki et al, 2003) (Lombeyda et al, 2001) † spans both GPU and CPU memory, along with a conventional preprocessed hierarchical data structure and hardware accelerated rendering to visualize large input data on output displays connected to distributedmemory clusters. Scientists successfully use an implementation of this approach, Vol-a-Tile 2, to visualize two large microscopy datasets, described later, on two LCD arrays.…”

Section: Data Sizementioning

confidence: 99%

See 1 more Smart Citation

Distributed Volume Rendering for Scalable High-Resolution Display Arrays

Schwarz¹,

Leigh

2010

Proceedings of the International Conference on Computer Graphics Theory and Applications

View full text Add to dashboard Cite

This work presents a distributed image-order volume rendering approach for scalable high-resolution displays. This approach preprocesses data into a conventional hierarchical structure which is distributed across the local storage of a distributed-memory cluster. The cluster is equipped with graphics cards capable of hardware accelerated texture rendering. The novel contribution of this work is its unique data management scheme that spans both GPU and CPU memory using a multi-level cache and distributed shared-memory system. Performance results show that the system scales as output resolution and cluster size increase. An implementation of this approach allows scientists to quasi-interactively visualize large volume datasets on scalable high-resolution display arrays.

show abstract

Section: Data Sizementioning

confidence: 99%

“…Parallel image-order techniques, also referred to as sort-first techniques, break the output image into disjoint regions and assign a processing unit to render everything in that region. Image-order implementations have been developed for sharedmemory systems (Palmer et al, 1998) as well as distributed-memory systems (Bajaj et al, 2000).…”

Section: Introductionmentioning

confidence: 99%

Distributed Volume Rendering for Scalable High-Resolution Display Arrays

Schwarz¹,

Leigh

2010

Proceedings of the International Conference on Computer Graphics Theory and Applications

View full text Add to dashboard Cite

show abstract

“…Palmer et al [31] studied two parallel partitioning and dynamic load balancing algorithms to explore the tradeoffs between their memory hierarchy performance. They suggest that image-order decomposition strategies suffer from a lack of locality in accessing the 3D volume data: during the ray integration loop -the most resource intensive part of the raycasting volume rendering algorithm -a given ray may need to access any voxel within the source volume.…”

Section: Previous Workmentioning

confidence: 99%

Multi-core and many-core shared-memory parallel raycasting volume rendering optimization and tuning

Bethel

Howison

2012

The International Journal of High Performance Computing Applica

View full text Add to dashboard Cite

Given the computing industry trend of increasing processing capacity by adding more cores to a chip, the focus of this work is tuning the performance of a staple visualization algorithm, raycasting volume rendering, for shared-memory parallelism on multi-core CPUs and many-core GPUs. Our approach is to vary tunable algorithmic settings, along with known algorithmic optimizations and two different memory layouts, and measure performance in terms of absolute runtime and L2 memory cache misses. Our results indicate there is a wide variation in runtime performance on all platforms, as much as 254% for the tunable parameters we test on multi-core CPUs and 265% on many-core GPUs, and the optimal configurations vary across platforms, often in a non-obvious way. For example, our results indicate the optimal configurations on the GPU occur at a crossover point between those that maintain good cache utilization and those that saturate computational throughput. This result is likely to be extremely difficult to predict with an empirical performance model for this particular algorithm because it has an unstructured memory access pattern that varies locally for individual rays and globally for the selected viewpoint. Our results also show that optimal parameters on modern architectures are markedly different from those in previous studies run on older architectures. And, given the dramatic performance variation across platforms for both optimal algorithm settings and performance results, there is a clear benefit for production visualization and analysis codes to adopt a strategy for performance optimization through auto-tuning. These benefits will likely become more pronounced in the future as the number of cores per chip and the cost of moving data through the memory hierarchy both increase.

show abstract

“…Various attempts have been made to parallelize rendering on a cluster of computers by image-order [37,2,3], object-order [13,16] or hybrid approaches [14]. The aim of these methods is to achieve a speed-up in the rendering process by combining the computational power of a set of computers.…”

Section: Related Workmentioning

confidence: 99%

Giga-Scale Multiresolution Volume Rendering on Distributed Display Clusters

Thelén

Meyer

Ebert

et al. 2011

Human Aspects of Visualization

View full text Add to dashboard Cite

Abstract.Visualizing the enormous level of detail comprised in many of today's data sets is a challenging task and demands special processing techniques as well as a presentation on appropriate display devices. Desktop computers and laptops are often not suited for this task because data sets are simply too large and the limited screen size of these devices prevents users from perceiving the entire data set and severely restricts collaboration. Large high-resolution displays that combine the images of multiple smaller devices to form one large display area have proven to be an adequate solution to the ever-growing quantity of available data. The displays offer enough screen real estate to visualize such data sets entirely and facilitate collaboration, since multiple users are able to perceive the information at the same time. For an interactive visualization, the CPUs on the cluster driving the GPUs can be used to split up the computation of a scene into different areas, where each area is computed by a different rendering node.In this paper we focus on volumetric data sets and introduce a dynamic subdivision scheme incorporating multi-resolution wavelet representation to visualize data sets with several gigabytes of voxel data interactively on distributed rendering clusters. The approach makes efficient use of the resources available on modern graphics cards which mainly limit the amount of data that can be visualized. The implementation was successfully tested on a tiled display comprised of 25 compute nodes driving 50 LCD panels.

show abstract

Ray casting on shared-memory architectures: memory-hierarchy considerations in volume rendering

Cited by 8 publications

References 21 publications

Distributed Volume Rendering for Scalable High-Resolution Display Arrays

Distributed Volume Rendering for Scalable High-Resolution Display Arrays

Multi-core and many-core shared-memory parallel raycasting volume rendering optimization and tuning

Giga-Scale Multiresolution Volume Rendering on Distributed Display Clusters

Contact Info

Product

Resources

About