Spectraseis Achieves High-Speed Imaging of Full Wave Microseismic Data
MAY 10, 2012, HOUSTON. Spectraseis, the technology leader in microseismic fracture imaging, stimulation evaluation and seismic monitoring, has achieved a substantial speed-up (compared to CPU implementations) for processing microseismic data, using proprietary algorithms and techniques to get the most out of NVIDIA® Tesla® GPUs based on NVIDIA's Fermi architecture.
The accomplishment provides faster turnaround times of both elastic and acoustic imaging of microseismic events for customers. This faster data processing method enables additional quality control and analysis of microseismic fracture monitoring datasets. Accurate event location enables asset teams to make more meaningful evaluations and planning of well completions and well spacing which leads to increased field productivity.
Spectraseis employs full elastic migration using 3-component field data that require the computational horsepower typically reserved for only the largest scale scientific models. To meet the challenge, Spectraseis has implemented optimized finite-difference propagation kernels on a cluster of compute engines, each containing seven NVIDIA Tesla M2070 GPUs providing thousands of separate compute threads, greatly improving turnaround times.
The size of the model domain, and relatively long recording times, create computational demands in processing data from fracture monitoring arrays. Reflection seismic processors have addressed comparable challenges with large hardware clusters, while microseismic analysis has to-date relied on many simplifying approximations to minimize the computational complexity, thereby degrading the accuracy and dependability of the data.
Spectraseis CEO Ross Newman said:
"Oil and gas operators are harnessing the power of GPU computing for applications such as seismic survey modeling and reverse time migration. Spectraseis has taken things further and implemented the full elastic wave equation for microseismic data on GPUs. Computation times measured in minutes and hours, instead of weeks, have now been achieved, even with complex imaging conditions. This allows higher quality full wave fracture mapping, using both pressure and shear data, to be delivered to customers in commercial timeframes. The benefit to our customers is a much better understanding of fracture performance and application of the most efficient frac parameters for future wells."
Sumit Gupta, Senior Director of the Tesla business at NVIDIA, said:
"The 28X acceleration Spectraseis is seeing in elastic wave equation imaging underscores the tremendous benefits GPUs provide to the oil and gas industry. NVIDIA GPUs are a true game changer in this industry, enabling companies to accelerate turnaround on major projects due to the availability of faster and more accurate results."
The elastic wave equation uses nine volumes of physical parameters that require constant interaction, resulting in many memory fetches. While an arithmetic operation is virtually instantaneous, a memory fetch uses 400-600 clock cycles. It is therefore critical to optimize both compute kernels and data management in the compute process. This is paramount for implementations across multiple GPU clusters, which require domain decomposition and message passing within propagation update calculations.
Spectraseis has tackled this problem by minimizing the number of memory transactions and aligning the data using an optimized patent-pending algorithm, resulting in a significant performance improvement. Additionally, a customized update/transfer overlap keeps the algorithm fast across multiple GPUs.