Layer | Fill | Outline |
---|
Map layers
Theme | Visible | Selectable | Appearance | Zoom Range (now: 0) |
---|
Fill | Stroke |
---|---|
Collaborating Authors
Results
NPML Boundary Conditions For Second-Order Wave Equations
McGarry, Ray (Acceleware Corp.) | Moghaddam, Peyman (Acceleware Corp.)
ABSTRACT We derive Nearly Perfectly Matched Layer (NPML) absorbing boundary conditions of direct applicability to second-order wave equations of interest in seismic modeling. Specifically, NPML equations are derived for the scalar acoustic wave equation and for VTI pseudo-acoustic wave equations. Extension of the formalism to other systems of second-order wave equations is straightforward. An efficient implementation scheme for the new equations on multi-threaded computing hardware is described. Finally, an example of the effectiveness of the new boundary conditions is presented.
Summary We give an overview of an implementation of Reverse Time Migration (RTM) on heterogeneous multi-core hardware based on Graphics Processing Units (GPUs). We demonstrate the clear advantages of GPU-based hardware for RTM, not only in terms of performance for small to mid-scale problem sizes, but also for imaging 3D volumes on a scale appropriate to full Wide-Azimuth (WAZ) or Rich-Azimuth (RAZ) surveys in areas such as the Gulf of Mexico. We also point to advantages of GPU-based hardware in terms of energy efficiency and datacenter footprint. Introduction RTM is a state-of-the-art technique for imaging subsurface geological structures that fully handles the two-way wave equation to improve imaging in areas of complex geology. The superior imaging quality of RTM comes at the cost of extremely intensive computational requirements which have traditionally limited its widespread adoption. RTM was first introduced in the 1980s (Baysal et al., 1983; Whitmore, 1983). The theory is well-known, so we do not dwell upon it here, other than to outline (in the next section) the two major barriers to implementing a practical RTM system for real large-scale imaging projects of interest today within the seismic industry. Despite the enormous computational requirements of RTM, it has recently been receiving renewed attention from the seismic imaging community, as evidenced by its increasing prominence in the relevant literature. This renewed interest has been fuelled both by a need for better imaging techniques and by advances in computer hardware, particularly with regard to processor speed and memory bandwidth. Otigosa et al. (2008) performed an evaluation of several high performance computing (HPC) platforms to assess their suitability for RTM. They considered both traditional CPU-based (homogeneous) hardware and "heterogeneous" hardware based on the IBM Cell/BE processor (see also Araya-Polo et al., 2009). These authors concluded that heterogeneous hardware does offer clear advantages for RTM over the more traditional CPU-only systems. In this paper we are also concerned with implementation of RTM on heterogeneous hardware, but our approach uses the Graphics Processing Unit (GPU) as a co-processor. Historically, GPUs were dedicated graphics rendering devices for a personal computer, workstation, or game console. The highly parallel nature of modern programmable GPUs, coupled with their exceptionally high memory bandwidth, makes them extremely effective general-purpose processors for a range of scientific problems. Each GPU contains up to 30 multiprocessors, each decomposed into a number of stream processors with generous on-board memory for user-managed cache. For example, recent NVIDIA GPUs such as the Tesla C870 and C1060 have totals of 128 and 240 streaming processors respectively. Each stream processor is fully capable of executing integer and floating point arithmetic. GPUs achieve high performance when thousands of threads execute concurrently, giving them a significant advantage over CPU-only hardware. The C1060 has 4 GB of on-board memory and a global memory bandwidth of 102 GB/s. In recent years GPUs have been successfully used in a number of scientific application areas such as airflow modeling (Fan, 2004) and electromagnetic simulation (Sypek and Mrozowski, 2008; Schneider et al., 2007; Krakiwsky et al., 2004). In the area of electromagnetic simulation in particular, GPUs are rapidly becoming an accepted technology.
- North America > United States (0.34)
- North America > Mexico (0.34)
- Energy > Oil & Gas > Upstream (0.87)
- Information Technology > Hardware (0.57)