Visualization on GPU Clusters and Smart Mobile Devices

Objectives

We are developing visualization systems that are capable of generating interactive visualizations of enormous datasets that result from physics based simulations on High Performance Computing (HPC) facilities. The visualization system will support both end user applications and a toolkit which enables the development of custom visualization tools. Using scalable rendering techniques, the system will operate on the datasets as they reside on the HPC platforms and display the results on the display units. Our target market is bio-science research labs., pharmaceutical companies engaged in drug discovery and molecular therapy, multimedia authoring companies using high definition clips, and companies doing CFD and CEM.

Vizag:

This product line is a cluster of GPUs based on Dual Xeon processors with state of the art Graphics Processing Unit (GPU) and optional accelerator boards. The PCIExpress based GPU nodes are connected using a one Gbps Ethernet switch. Cluster configurations are customized by users. The software for the cluster is also customized by users. The system will be built based on customers order; selected software will be installed and tested using specified applications before shipping.

Vizag-HDA:

This product line is also a cluster of GPUs based on Dual Xeon processors but the software is tailored towards High Definition (HD) capturing, editing, and authoring. The PCIExpress based GPU nodes are connected using a one Gbps Ethernet switch. Cluster configurations are customized by users. The software for the cluster is also customized by users. The system will be built based on customers order, software installed, and tested using selected application before shipping.

Background

The simulations in science and engineering regularly generate terabytes of data that needs to be analyzed via scientific visualization techniques. Moving that data to a remote location for visualization is impractical if not impossible. While techniques such as feature extraction and sub-sampling can be useful, it is commonly necessary for important visualization algorithms to operate on the entire dataset. Performing these algorithms on the HPC platform itself is a natural step in this process. However, during the past decade, the increase in the power of commodity graphics processing unit (GPU) cards has outpaced the native, integrated graphics sub-systems available on HPC platforms. Groups of inexpensive commodity GPU systems are used to construct multi-display systems. By combining the output of multiple displays via rear-projection screens, scalable display resolutions have been achieved. The challenge in the next decade is to make visualization possible on smart mobile devices. Smart phones already have support for 3D graphics and games. Visualization of scientific and simulation data on smart mobile devices will open up new applications in science, medicine, factory automation, and patient care. The Vizag line of products are aimed at this market.

We are also developing systems that could visualize structured and unstructured grid datasets using polygon-based and volume visualization techniques using HPC platforms for the algorithms and single or multiple smart mobile devices (SMD) for the display and interaction.

Structured grid dataset size can reach the multi-billion cell level while unstructured grid datasets are typically an order of magnitude smaller. Displays can be a single SMD or a multi-display co-planar system (DisplayWalls). The system would need a graphical user interface as well as a scripting interface for custom visualizations. While a single visualization application may be sufficient for many simulations, there will always be a need for customization. Therefore, the system should be capable of being utilized from a SMD using scripting languages. This will allow both the production visualization environment and the customizable tools to be based on the same subsystems.