At the Super Computing 2007 show in Nevada last month, Mercury Computer Systems was showing a prototype of graphics processing unit (GPU) based number crunching from inside Schlumberger’s Petrel interpretation flagship. Mercury’s Open Inventor-based 3D visualization engine (a component of Petrel and Schlumberger’s Ocean development platform) has been interfaced to NVIDIA’s Tesla high performance computing (HPC) engine to address the computation and visualization of large 3D seismic data sets.
The resulting ‘Probe’ technology was also on view at the SEG last month where the HPC toolkit was used in direct reservoir identification and for ‘WYSIWIG’ geobody extraction and editing (OITJ October 07). Ocean embeds Mercury’s VolumeViz LDM high-performance volume visualization technology for large seismic data sets.
The new technology results from a deal between Mercury and NVIDIA to add the GPU-based ‘CUDA’ code set to the Open Inventor toolkit. This effectively opens up general purpose GPU (GPGPU) HPC to all users of the Ocean development kit.
Mercury VP Jean-Bernard Cazeaux said, ‘3D visualization has revolutionized the understanding of seismic data, thanks to the performance provided by the GPU. But now GPUs do more than visualization, they provide amazing computing capabilities for interactive applications.’
GPU-based computing is showing promise in HPC because the number crunching capacity of graphics chips is fast outstripping that of the conventional CPU (Moore’s law). Tesla, which comes in several form factors from a PCI Express card to a rack mount, is a high-performance hardware device that adds supercomputer power to a CPU-based workstation. The CUDA C-Compiler allows optimized development on the GPU using the C programming language.
The CUDA/Tesla combo provides application developers with access to high performance seismic processing capability with simultaneous visualization of results in 3D. Andy Keane, General Manager of the GPU Computing business at NVIDIA added, ‘Some geophysical tasks necessitate supercomputer power. By providing this on a standard workstation, we are enabling a fundamental transition in the way you define your workflow. By reducing some computation times days to hours, NVIDIA Tesla is a significant disruption to HPC.’
© Oil IT Journal - all rights reserved.