• The system achieves unprecedented speed in visually rendering large-scale volume data, including data generated in numerical fluid simulation and 3D medical imaging.
• Using this technology, our researchers have developed “frame compositing hardware” for parallelization of high-performance PC graphics processors and a “volume graphics cluster system”.
• These advances in the development of large-scale numerical computing now enable instantaneous visual monitoring, resulting in huge gains in operational efficiency.

The Collaborative Research Team of Volume Graphics at Advanced Industrial Science and Technology (AIST), an independent administrative institution, working in cooperation with Mitsubishi Precision Co., Ltd. (MPC), has developed a “volume graphics [VG] cluster system” that adds real-time visualization capabilities to conventional PC cluster systems.

While there has been a recent increase in the use of PC clusters for numerical fluid simulations, 3D medical image processing, and other applications requiring processing of large-scale volume data, the technology had not reached a level at which the results processed from such massive amounts of volume data could rapidly be converted to visual images. AIST and MPC now have developed “frame compositing hardware” in which multiple high-performance PC graphics processors operate in parallel, successfully clearing the way to commercial production of VG systems that can handle both computation and visualization of large-scale volume data in real time.

As this technology permits instantaneous visual monitoring of large-scale numerical calculations while computation is in progress, it is expected to bring huge gains in operational efficiency in fields such as numerical fluid simulation and 3D medical imaging.

Meanwhile, MPC plans to initiate sales of VG cluster systems and related products while continuing to collaborate in the development of VG cluster technology for use in every field.

Recently, the technology known as PC clustering has been attracting attention the world over. By installing the Message passing library for parallel computers developed by Argonne National Laboratory and Mississippi State University (MPICH) or similar software on large numbers of personal computers (PCs) connected over an ultra-high-speed network, nearly anyone can create an extremely low-cost parallel computer with performance surpassing supercomputer benchmarks. If the PC cluster is used only for performing numerical calculations, though, the resulting data must be transferred to a separate workstation for visualization.

In contrast, if each PC’s graphics processing unit (GPU) is set up so that it functions in parallel with the others in processing visualizations, the PC cluster is then able to perform both numerical computing and visualization simultaneously, dramatically increasing the efficiency of data processing. However, parallelization using conventional networks and slower central processing units (CPUs) has not sufficiently brought out the potential of each GPU.

With the use of PC clusters being limited to universities, research centers, medical institutions, and other higher-level research facilities, manufacturers face significant economic risk in independently researching problems in these fields. In light of this issue, AIST, with its expertise in VG technology, joined with MPC, a developer of various types of simulators and real-time imaging devices. Provided with funds budgeted for the “Visual Computing Technology for Real Time Bio-Information Processing” project of the former Agency of Industrial Science and Technology and funds supplied though the AIST Collaboration Department’s Industrial and Academic Collaborative Research System, the group developed its VG cluster.

Two approaches are used in parallelization of graphics processing: one in which images are shown on multiple displays, and another in which the target object (data) is processed separately. Our VG clusters, however, combine both methods through our newly developed frame superimposition system. The system is capable of taking the output of each PC’s GPU at rates of more than 1 gigabyte per second, keeping track of target object color, opacity, and back-to-front relationships, while simultaneously outputting the synthesized images. Combining this with the PC cluster’s number crunching capabilities fully utilizes the system’s power in real-time simulation of numeric fluid dynamics, bio-physics, and other applications.

Although the speed at which images are generated may vary according to the amount of data being processed and the size of the displays used, the VG cluster system’s high degree of scalability means one can fashion a hybrid, real-time computer visualization system that can incorporate image-space and object-space subdivisions according to the size of the problem.

While continuing to collaborate in developing VG cluster technology for use in a variety of fields, the group is also developing systems in which VG cluster sites are parallelized over wide-area networks, while further investigating the development of systems using multiple screens and immersive displays for virtual reality and other applications.

Furthermore, MPC is preparing for sales of VG cluster systems and frame compositing hardware kits to universities, research institutions, and other such organizations.