GPU-Based High-End Desktop Computing, A Cost-Effective Solution to Improve Reservoir Simulation Performance

Abstract

Abstract The ever-increasing resolution of reservoir simulation models, combined with the emergence of ensemble approaches, call for the adoption of simulators running on Graphics Processing Units (GPUs), as these offer unprecedented performance on limited hardware footprint. We here discuss how the adoption of such a tool within Mubadala Petroleum allows engineers to seamlessly run partners' models on local workstations, avoiding the complications of accessing cloud computing resources (no data transfer authorizations, challenging cost-control, risk of vendor lock-in, etc.). Adopting new business-critical tools such reservoir simulators require careful validation; not only is the performance important, but also the stability and reliability. For instance, are the results sensitive to minor changes in the model? Or some others can lead to a change in hardware? In effect, in any nonoperating holding-company, it is furthermore important to be able to masterly reproduce the documented profiles. After a review of the principles of GPU based reservoir simulation, the process undergone to perform the above-mentioned validation were presented in this paper using local workstations. Different simulation models with different levels of complexity have been tested using the GPU computing architecture of mid-end and high-end desktop computer (MEDC and HEDC). History matching and prediction runs were carefully validated against the original benchmark results. Additional tests and the performance scalability across various configurations were also explored. The comparison of different GPU-based computer gave the appropriate results compared to a highly invested option of having multiple multi-core processors, a cluster, or a networked computer. The use of GPU-based high-end desktop computer (HEDC) will allow to deliver recommendations in time to support final investments decision, primarily because parallel solution simulations speed up simulation time, which is especially important in carrying out prediction solutions scenarios. From this perspective, the added value of using GPU's HEDC presented in this study shows the significance and most convenient solution for decision making as a suitable option for Mubadala Petroleum.