What do video games and seismic exploration have in common? Both require very demanding computer applications that call for the ability to process massive quantities of data rapidly.

A consortium called the Mission-Oriented Seismic Research Program (M-OSRP), based at the University of Houston, has joined forces with IBM to examine the use of computer technology originally designed for video-game consoles to process massive quantities of data rapidly.

This may be a new toy for young geoscientists, sometimes called “joystick geophysicists,” who are more comfortable with video games than geologic maps. If these two technologies merge, video games may be training the next generation of geoscientists. The consoles the games are played on may very well be the format for the next generation of processing and interpretation tools.

“IBM is supporting [M-OSRP] with a Cell Broadband Engine (Cell/BE) system that represents a new generation of powerful supercomputers with substantial parallelism built in at the core level,” the consortium reports. “Such highly parallel computing technology is characterized by multiple processors executing and analyzing different types of data at once.”

The Cell/BE processor was originally designed for the Sony PlayStation 3.

The idea behind M-OSRP is to change the way geophysicists think about seismic-processing algorithms, and to do this effectively it needs better computing power.

“One of the algorithms developed within M-OSRP to suppress a form of coherent noise called internal multiples places a high bar on seismic data collection and a very high bar on computing speed and memory. To allow the petroleum industry to use this very effective methodology for 3-D data will require a new computing vision and capability.”

IBM researchers have recoded the algorithm for the Cell/BE processor and are running comparisons with industry-standard computer architectures.

Meanwhile, another team chose to simply crunch the data on the PlayStation itself. In “Playing with a seismic PlayStation,” C.J. Bednar and J. Bee Bednar of Panorama Technologies presented their findings at a recent European Association of Geoscientists and Engineers annual meeting in London.

“Rather than focus on only one computationally intensive algorithm, we attempt to discuss various aspects of the more popular seismic-imaging algorithms,” they report. They go on to discuss the nature of software optimization in the Cell and memory limitations relating to ray tracing, Kirchhoff imaging, common azimuth imaging, one-way dual-domain imaging, full finite-difference two-way reverse time imaging and seismic modeling. They also compare the computational speed of the Cell processor with modern multi-core chips from AMD and Intel.

The authors draw a preliminary conclusion: “Assessing the difficulties associated with achieving acceptable performance on PS3-style devices suggests that programming them will not be a serious impediment to programmers already familiar with high-performance computation and I/O code.

“[The] coding is not much more challenging than posix-threads programming, although it is also necessary to vectorize and optimize for the memory structure.” They add that Panorama Technologies’ entire software suite was installed without difficulty.

Various modeling and imaging techniques were easily modified to take advantage of the Cell architecture and it is not beyond imagination to think the system could operate 20 to 25 times faster than equivalent 2.2 GHz processors, they report. However, initial tests on ray tracing did not achieve “acceptable performance,” despite the fact that ray tracing is one of the primary aspects of any 3-D gaming application.

Comments