Oil and gas discovery poses a number of challenges for high-performance computing and visualization. As the ability to gather detailed and precise geologic data grows, so does the size of the datasets that need to be analyzed. This results in the need to process and visualize ever-growing amounts of data. Fortunately, applications that take advantage of the parallel computing power of the graphics processing unit (GPU) make working with such multiterabyte datasets in real time possible and affordable.

The importance of fast processing of geophysical data cannot be understated. In addition to
 
Figure 1. Showing crossline and inline poststack data together with associated prestack data (small jutting sections) on both crossline and inline. To the right is a prestack probe (prestack data volume) which is using a custom shader to render uninteresting data transparent while emphasizing areas of interest. (Images courtesy of Headware Inc.)  
the ever-growing datasets, another factor is the need to examine more locations for the possibility of oil and gas deposits. As the search goes further afield, more data has to be analyzed. And as data acquisition methods become more sensitive, separating the valuable grains of data from the chaff of the noise requires greater processing power.

A new generation of software and hardware tools are coming online that bring the superior data processing power needed for these tasks. One of these is from Headwave Inc. Headwave enables real-time visualization and analysis of prestack and poststack data; optimizes data for efficient workflows; and allows for real-time, remote collaboration between analysts and interpreters.

Headwave

Headwave’s solution allows geophysicists to apply advanced filters to their data and instantly visualize results even on multiterabyte prestack datasets. In addition, with Headwave’s software, geophysicists can analyze the original acquired prestack data in multiple dimensions as part of their daily workflow. The ability to visualize the entire prestack survey in real time is a powerful capability that until recently was not practicable on available computer systems.

The platform presents seismic data in multiple ways (shot or common midpoint [CMP] gathers, CMP volumes in either 2-D, 3-D or 4-D) and can switch between different domains. Analysts can use probes or planes to examine additional data, and prestack data can be linked intuitively to its corresponding poststack data.

The prestack and poststack datasets
are stored in a multidimensional, compressed format that also preserves header information and metadata for use by other applications in the workflow. The GPU in the workstation can be used to decompress the data on the fly.

Normally, the access of arbitrary parts of terabyte-plus prestack data in real time would
 
  Figure 2. Another probe of the prestack and poststack data. There are four prestack sections (vertical stubs) and multiple prestack time layers (horizontal); there is also a prestack probe. In this case the standard shader has been used, and thus the prestack volume is opaque.
require hundreds of cluster nodes. However, the Headwave engine can handle this on a single PC workstation, even at full 32-bit resolution.

The algorithm for wavelet-based compression and decompression of seismic data is highly optimized for maximum performance and makes extensive use of the GPU for computational tasks associated with compression/decompression. Implemented on a CPU, the wavelet compression/decompression algorithm achieves rates of about 10 MBps. But by exploiting the parallel processing capabilities of the GPU, the company is able to achieve rates of more than 2 GBps — a performance speed-up of 200 times. Other algorithms in the package run from 10 to 70 times faster on the GPU than on the CPU.

In addition to making use of GPUs for computation, Headwave has an optimized data processing pipeline to achieve as high a performance as is possible. The company has also achieved significant efficiencies in intra-node data transport, data caching and retrieval, and it incorporates sophisticated load-balancing tools to enable efficient scaling of the solution from single workstations to large clusters.

Improving workflow further, Headwave enables remote visualization of the prestack data so that geophysicists can work from their desks instead of moving their worksite to a visualization center.

GPU computing
The casual observer might expect GPUs to excel at visualization tasks, but what is not self-evident is that the architecture of the GPU makes it ideally suited to parallel data processing as well. Graphics rendering is compute-intensive, highly parallel computation — the same functionality that is required for many areas of geophysical data analysis. Unlike a CPU, a GPU has more transistors devoted to data processing than to data caching and flow control. And with up to 128 processors and memory bandwidth of up to 76.8 GBps, the latest generation of GPUs offers extremely cost-effective computing power for both visualization and computing applications.

The GPU is especially well-suited to address problems that can be expressed as data-parallel computations with high arithmetic intensity — in other words, when the same program is executed on many data elements in parallel with a high ratio of arithmetic to memory operations. To give a metaphorical example of parallel processing, if you had to search a book for the appearance of a particular word, a CPU would start at page one and progress sequentially through to the end of the book. A multicore CPU would divide the book into two or four sections and search each simultaneously, arriving at the result somewhat faster. A GPU, on the other hand, would divide the book into thousands of sections, achieving the result orders of magnitude faster.

While a CPU uses a single processing program to loop over data sequentially, data-parallel processing with a GPU maps data elements to thousands of parallel-processing threads.
Applications that process large datasets such as arrays or volumes can use a data-parallel programming model to accelerate these computations. This acceleration is further increased with the even more powerful GPUs that have recently come to market and by the fact that most workstations can now support multiple GPUs, either directly through PCI Express slots on the motherboard or via deskside visualization systems like NVIDIA’s Quadro Plex and NVIDIA Tesla for even more processing power.

Until recently, wide-scale use of GPUs outside of graphics applications has been inhibited by the lack of a practical software development environment and the necessary systems for customers to easily use GPUs in their computing work. But with the introduction of the NVIDIA CUDA software architecture, applications can be programmed in C to make use of GPUs, obviating the need to program in OpenGL, which is not well-suited for non-graphics applications. The ability to use standard programming tools and more easily convert existing applications to use the GPU will further speed adoption of GPU computing by the industry.
Oil and gas company workstations are already typically outfitted with NVIDIA GPUs, meaning that much of the hardware infrastructure to exploit this technology is already in place. Where additional systems or system upgrades are required, the new acquisitions will fit seamlessly into existing IT infrastructures without investment in custom systems and new types of technology or extensive retraining of IT personnel.

As a result, oil and gas companies are well positioned to start taking advantage of Headwave’s technology.

Oil and gas companies will increasingly be able to combine ever larger datasets, higher resolution visualization and more raw computing power to make discovery of new reserves faster, easier and less costly.