Game for advancement
  from: Offshore Engineer
  by: Steven McGinn
  Wednesday, April 09, 2008

Splicing hardware for computer gamers with the requirements of the E&P;and seismic industry seems an unlikely marriage, but demand for high-end visuals and graphics in games fueled the advancement of hardware for data processing. Steven McGinn reports on this and other recent IT-related developments in the offshore sector.

Programmers realized nearly five years ago that when coded correctly, graphics processing units (GPUs) can process data. Instead of clusters of supercomputers and rooms of networked servers, seismic analysts can access their data in a smaller, more energy efficient fashion through GPUs.

‘The demand for high-end hardware came first, all driven by the computer gaming market,’ reports software development company Headwave’s vice president of integration and deployment Steve Briggs. ‘And now companies that process massive amounts of data are beginning to use the GPU architecture instead of a computer’s central processing unit.’

Processing visuals in games is a matter of computing large amounts of data, rendering that data into an image, and then displaying the image on screen. Given that the nature of graphics relies on simultaneous multi-threading for processing complex algorithms into images, GPUs are likely candidates for scientific computations.

‘At the core level, GPUs are far more efficient at dealing with parallel computing than general-purpose processors,’ says NVIDIA GPU computing general manager Andy Keane.

Before 2002, Keane says, data computation on a graphics card was primitive because the card still thought it was processing visuals, not data. The hardware needed to be reprogrammed and ‘told’ what it needed to do.

‘By tricking a video card into thinking that it is processing data, you can make its performance substantially faster and more efficient than a regular processor,’ explains Briggs.With the increasing programmability ability to ‘trick’ standard GPUs, graphic chips can perform more than the specific graphics computations for which they were designed. In the past, big machines were used for computation and rendering, with clusters of PCs being adopted for such tasks.

Because GPUs are capable co-processors, their high speed makes them useful for a variety of applications, says Keane, not least as an alternative to networking together rooms full of computers for future seismic dataprocessing requirements. ‘We started programming and executing data on graphics boards and found that they perform faster because graphics cards are built with 128 clusters behaving as a computing chip for processing data,’ adds Keane.

Designing application data is off the beaten path for NVIDIA, a company better known for the 3D graphics chips found in the computers of media enthusiasts and gamers. But in their quest to develop a way of processing large amounts of information from dataintense applications ranging from graphics rendering to medical research and data farming, NVIDIA’s engineers came up with the Tesla product line. It offers three solutions: a dedicated board inside a personal computer, a desk-side computer, or a rack-mounted server attachable to a network’s data center.

‘It’s like your own personal supercomputer,’ Briggs says. ‘Ten years ago, the single (one-up) 1U S870 with four cores would have ranked number one on the list of the world’s most powerful computers.’ Each GPU chip on Tesla hardware has 128 processors running multiple threads, with 10,000 simultaneous threads under hardware control.

Keane says NVIDIA developed Tesla by hiring software engineers to work alongside the programmers and hardware specialists, so by default the architecture had to be written to work with the hardware.

‘We had to make a way to program the GPU and have the right features,’ he notes. The reason, he says, is that the data-hungry users wanted an expanded floating point power, but those types of configurations of data are often slow. NVIDIA’s response was to create the Compute Unified Device Architecture (CUDA) C-language for Tesla hardware. Alternative chip architectures, such as the GPU, provide higher floating-point performance than conventional CPU chips, but require even more unconventional programming models. CUDA gives computationally intensive applications access to the processing power of GPUs through a programming interface, enabling the GPU to solve complex computational problems. Providing orders of magnitude more performance and simplifying software development by using the standard C-language, NVIDIA says data on the Tesla hardware can be processed at a rate of 450 gigaflops a second per chip with 170 watts of power. A CPU processes data at 15 gigaflops a second per chip at 80 watts of power. If an analyst is using a quad-core CPU, producing between 60 gigaflops and 80 gigaflops, using a Tesla C870 is nearly four times as efficient, says the company.

Mercury Computer Systems developed VolumeViz, another GPU-based architecture solution for data processing and framework for managing computations and renderings. Its technical director, Michael Heck, says VolumeViz uses hierarchical multiresolution bricking, which initially subdivides the volume data and loads the lower-resolution information. After refining the low-res data, the application loads high-res information into the background, so the user doesn’t have to wait for all the data to be loaded. According to Heck, the single machine is a viable platform for visualization of E&P;data.

Not all seismic analysts, however, will turn to GPU architecture or a singlemachine solution for data analysis. Among reported drawbacks for Tesla are that the coding has to be specifically written for the hardware, and that it only solves a certain type of mathematical problem. ‘If we want to use the new graphic processor, we have to change the code,’ says Chen Bin Su, CEO of TGS-Nopec subsidiary Parallel Data Systems. They will, he adds, stick to cell processor technology of the kind found in Sony’s PlayStation3.

According to Su, it just made more sense for his company to stay with cell technology. ‘Tesla uses array processors for coding and the human investment in the software coding is not as important for us as the hardware investment. There really is nothing wrong with what we have.We just look for a faster and cheaper way to process volumes of prestack data, and cell technology is superior for us,’ says Su, adding that parent company TGS-Nopec tested Tesla three months ago. ‘If we stick to cell processor, we don’t have to change a thing,’ Su concludes. OE

New releases

Recent software releases expected to find ready application in the offshore sector include an updated version of CloudWorx from Leica Geosystems. For AutoCAD users working with 3D point cloud systems such as high-definition surveying, the new version, 3.3, incorporates a clash detection manager, a flange point selection tool, and a point cloud density display control. Leica says its point density display control will speed up the process when working with large data sets.

‘The point cloud clash management functionality allows plant designers to readily find clashes between existing “as-built” conditions and the new piping, equipment and structural designs inside their Plant CAD application,’ says Bill Wallace, senior software product manager for Leica Geosystems HDS. ‘The flange point selection tool enables the definition of tie-in points, and the point density display control capability speeds up office productivity. The company says such data sets could otherwise take a long time to display on screen before allowing the user to focus on a selected area.

Meanwhile, the latest releases from Strucsoft Solutions include two software options for offshore fabricators: PVGen Pro for topside pressure vessel manufacturers, and Jacket Pro for offshore substructure builders.

PVGen Pro can generate a 3D model of any pressure vessel by reading its geometry and specifications from the equivalent compress file. With the included library of connections with PVGen Pro, users can edit or add steel items to the pressure vessel, such as ladders, platforms, handrails and catwalks.

In Jacket Pro, jackets, barges, lifting booms and other steel structures can be automatically generated by inputting a few parameters, such as section sizes, elevations, and material type. The application generates 2D drawings, including any combination of sections, views, and parts, and it automatically updates the drawings if the 3D model changes.

According to Strucsoft, both software applications will improve productivity by producing multiple results from the same efforts. Both sets of software include direct machine output through DSTV format, CNC output, and communication with plate cutters via layered .dxf files.

Primavera meanwhile has released the latest version of its project risk analytics software for project and portfolio management environments. According to the company, version 8 of its PertMaster software includes a new cost/schedule trade-off analysis showing the relative benefits of alternative risk mitigation approaches, a new interface, extended integration with Primavera P6 and Microsoft Project 2007, and a standard risk assessment function.

‘Primavera PertMaster v8 has given my team even greater insight and ability to ensure that the project keeps moving to meet with time and budget constraints,’ says Bill Hunter, Aker Kvaerner’s manager for planning and scheduling. ‘The speed of the information back to the risk management team makes it easier to respond to challenges that arise.We can also alert management so there are no surprises.’

Making waves with prestack data

Headed by former WesternGeco geophysicist Alex Krueger, Houston-based software developer Headwave offers oil and gas industry 3D visualization software designed specifically for viewing and interacting with prestack data.

By allowing for interaction with multi-terabyte prestack surveys, Headwave says it offers users a substantial leap in time and effort of processing huge sums of data. ‘With this compelling new GPU architecture from NVIDIA, we can accelerate some of the most computationally intensive algorithms in oil and gas exploration, far beyond the performance CPUs are capable of delivering,’ explains Krueger.

His company uses GPUs for raw processing power, providing users with a unified view for enormous amounts of data and applying geophysical algorithms to the datasets with real-time results for interactive visualization.

Whereas it traditionally saw 3D visualization as a tool for the interpreter and the reservoir analyst, Headwave now believes the features that make it such a powerful tool for poststack analysis are directly applicable to prestack workflows.

Agbami zooms in

Operator Chevron selected engineering IT solutions supplier Aveva to provide and implement the Aveva NET Portal program to manage critical information for the operation of the Agbami FPSO off Nigeria.

Equipped with the latest environmental protection technology, Agbami will be located 70 miles offshore Nigeria and will produce 250,000b/d while injecting water and gas for reservoir pressure management (OE January).

Chevron will manage the FPSO from its offices in Lekki, Nigeria, with associated design and engineering work coordinated over a number of sites internally in Nigeria as well as the offices of a number of international contractors. internationally and on the FPSO itself.

The Aveva NET Portal continually updates information on a master hub and synchronizes it with replicated data onboard the FPSO, which will include the 3D facility model, documents, and other information ensuring accurate data is available online at any time during operation.

Click here to register to receive your own copy of Offshore Engineer each month.