Optimize Computational Throughput of Reservoir Simulation Workflows in a GPU Based High Performance Computing Environment

Abstract

Abstract Decision making workflows in the energy industry require the capability to simulate hundreds if not thousands of models in a practical timeframe. In this paper we show how these capabilities can be achieved by exploiting top-class high performance computing resources available on premise. A GPU reservoir simulator can maximize the efficiency and effectiveness of computational kernels for solving partial differential equations and phase equilibria on modern devices, hosting thousands of floating-point units with an unprecedented memory bandwidth. It incorporates functionalities to model recovery processes, and includes capabilities for coupled reservoirs and network simulation. Recently, we added the possibility to exploit GPU Multi Process Service and run field scale, multiple-reservoir simulations on a single device. The Simulator enables the possibility to simulate multi-million cell models on a single GPU, and then execute reservoir workflows where ensembles of many hundreds of models may run altogether, maximizing the overall throughput. In this work we present the most relevant applications of a specific GPU reservoir simulator, highlighting its efficiency and effectiveness in tackling everyday business requests and the benefits it provides with respect to traditional CPU-based software. In more detail we will show the application of the tool for the simulation of: 1) an uncertainty quantification of a compositional dual porosity-dual permeability super-giant carbonate reservoir; 2) the history matching of a black-oil reservoir through ensemble multiple data assimilation followed by screening and optimization of injectors location; 3) revisiting an integrated asset model uncertainty quantification including five black-oil reservoirs. This work demonstrates how the availability of a GPU centric reservoir simulator combined with high performance computing resources was instrumental to the deployment of the most complex and computationally intensive workflows processes inside our Company. The process combines focus on the leading-edge company hardware and care on current company needs.