In the last decade the evolution of technology which allows geoscientists and engineers to visualize large volumes of 3-D data has been one of the great advances in oil and gas exploration and production (E&P) computing, acknowledged as such by industry leaders.

Three-D visualization has played a critical role in improving prospect evaluation from seismic

Figure 1. VGP can optimize the use of multicore CPUs and improve their rendering performance and scalability. (All figures courtesy of ModViz)
data by helping to better understand the complex geometries of the subsurface and spatial data relationships. Interpreters and modelers have benefited in their ability to handle large and increasingly complex volumes of data and associate them with rock and fluid properties.

Today we find that the graphics computing needs of the E&P business have reached a new tipping point. The large-scale symmetric multiprocessing (SMP) systems which epitomized the first visualization revolution still have their place, but they are expensive and offer only limited access to a potentially large user group in an oil or service company with offices around the world. The focus now is on leveraging the latest computer hardware to radically improve the visualization possibilities for individual workstation users and application developers. The goal is to provide supercomputing visualization performance on the desktop and, in doing so, to enhance the effectiveness of a much wider community of interpreters and model builders in the industry worldwide.

Seismic data is extremely complex, in part because of the multiple attributes per data point and the graphical diversity with opaque and transparent surfaces, lines, points, 2-D and 3-D transparent volumes, etc. When interpreting these data, geoscientists and engineers not only need to visualize but also need to be able to manipulate, interact with and edit them. This can be an extremely graphics-intensive task.

Multiple datasets and attributes are involved in typical E&P workflows so that even a simple workflow can easily take up the full capability of a high-end graphics card. For example, rendering large seismic volumes with variable opacities is important in finding fluid contacts, but the results are limited by the rendering performance possible and by how much data can be processed each time. Today’s state-of-the-art graphics cards (e.g. NVIDIA Quadro FX 5500) support almost 1 gigabyte of memory available to store 3-D texture data, but this memory can easily be exceeded by the smallest of geophysical datasets.

The result is that geologists and geophysicists end up subdividing their work and interpreting only a portion of the dataset at a time, thereby reducing their productivity and limiting their ability to see the “bigger picture.” This scenario has been described as trying to view a landscape through a soda straw; that is, one can only examine a very small area at any given moment.

The workstation visualization challenge for E&P is characterized by computationally expensive algorithms, a very large number of diverse datasets, and a need for greater interactivity and collaboration. Being able to process, visualize and interact with a large amount of data in real time is the key for improved interpretation and reservoir modeling workflows.

Part of the solution to these issues was developed around multicore processors, which represented a major evolution in computing technology. Placing two or more powerful computing cores on a single processor opened up a world of new possibilities. In very recent memory the use of multiple central processing units (CPUs) and multi-cores accelerated the computational processing speed of machines and, for example, facilitated the development of computer-intensive seismic data processing techniques such as prestack depth migration, which is now commonplace in the industry.

Figure 2. An unmodified GoCAD interpretation is shown on the left, with its spatial decomposition shown on the right.
The more recent big step forward has been the development of multiple graphics processing units (GPUs) which dramatically accelerate graphics processing speeds and increase the graphics memory capacity. The implications for visualization workflows on the desktop are something to get excited about. Geoscientists and engineers can now be empowered to generate a more comprehensive visual story from field data to geologic model to drilling completion, effectively capturing and understanding the transformation of subsurface field signals to rock and fluid properties.

The graphics revolution stems from the new generation of workstations which have two or more graphics boards, high-bandwidth PCI Express connections, and four CPU cores. At the Siggraph computer event this year, NVIDIA introduced a dedicated visual computing system (Quadro Plex VCS) that can be easily connected to a workstation or a graphics server, allowing up to eight graphics processors. This contrasts with the single graphics cards still found in many workstations in the industry where there has been no recent technology refreshment.

The interactivity and performance required by the E&P industry can be achieved by combining the power of multiple processors and graphics cards that are commonly available in the latest workstations. However, multiple CPUs, CPU cores and GPUs make programming more complicated for software developers. They have to implement complex methods to distribute work among GPUs and CPUs in order to achieve successful scaling.

Figure 3. Interacting with four times the amount of data using VGP-enabled Shell 123-DI interpretation software on a standard workstation equipped with four graphics processing units.
ModViz specifically developed its Virtual Graphics Platform (VGP) to tackle this complex problem and to take advantage of multiple CPU cores and GPUs for advanced 3-D visualization, allowing application developers to focus on their application goal, not on how to achieve graphics scalability.

The benchmark example in Figure 1 shows how VGP can optimize the use of multi-core CPUs and improve the rendering performance and scalability by transparently virtualizing 3-D graphics-intensive applications across multiple GPUs and multiple CPU cores. The objective is to improve human efficiency at all stages of the E&P workflow.

ModViz began development of VGP 21¼2 years ago to provide a software layer that would be transparent to the user as it managed CPU cores, system memory, GPUs and the graphics memory for an application. In practice VGP intercepts the application’s OpenGL data stream and analyzes and intelligently subdivides and distributes the optimized portions of the OpenGL data to an array of GPUs available in the operating computing system for parallel processing. It then synchronizes, composites and delivers the resulting images back to the user’s application or out to external displays. VGP has applied a number of high-performing techniques, among them OpenGL stream optimization, load balancing, real-time data management and distribution, high-speed compositing, and shared memory optimization, all with the intention of minimizing and streamlining communications between various CPUs and GPUs. All this is done transparently to the user’s application and in real time.

The result is a highly efficient parallel rendering system enabling E&P geoscientists and engineers to work with much larger 3-D seismic data at a much faster speed while using their existing applications.

The computing resources in today’s workstations were only available in a commodity cluster 2 years ago, considering the number of cores, number of GPUs and the amount of system memory available. The PCI Express bus has provided workstation builders with an extremely fast, bi-directional interconnect which has helped drive the increase in computing resources inside the box.

For the same reason, VGP was initially introduced on clusters of workstations because this was the only way to obtain access to an array of GPUs other than via specialized and high-priced SMP systems. The launch of multicore processors and multi-GPU hardware has not only “commoditized” the high-end SMP systems but also made it accessible for mainstream users. It also presents a much more compact and optimally designed hardware for VGP to work with.

VGP Version 2.0 was announced at this year’s Society of Exploration Geophysicists meeting to put supercomputer-level visualization in the grasp of E&P workstation users. The software can also be applied to the engines driving large visualization centers. A number of oil companies, notably Shell, BHP Billiton, ConocoPhillips and Nexus Geosciences, have installed VGP. In the case of Shell, the company is deploying VGP globally to accelerate the performance of its proprietary 123-DI 3-D seismic interpretation application (Figure 3).

Comments