Use of Graphics Processors in Calculation of Full-Scale Thermal Models

Advancements in computational technology have opened new possibilities for simulating complex processes in the oil and gas industry, including thermal enhanced oil recovery (EOR) methods and carbon capture and storage (CCS) techniques (Bui, Mai et al., 2018). One of the most commonly used models in this field is the thermal three-phase compositional model. These models require extensive computational resources, as they must account for multiple components (hydrocarbons, water, CO2) and their interaction across various phases (liquid, gas, and solid). The high resolution of hydrodynamic grids, the large number of components, and the highly nonlinear dependence of thermodynamic properties on temperature impose significant demands on computational resources. Accurate modeling of real-field processes necessitates high-performance computing systems with substantial memory bandwidth. With the development of graphics processing units (GPUs) and supporting software, high-performance computing systems increasingly leverage GPUs for such simulations (Nickolls and Dally, 2010). However, to fully utilize GPU capabilities for thermal modeling calculations, traditional CPU - based algorithms must be substantially restructured. This paper presents a hybrid GPU + CPU implementation of the computational core for a thermal compositional reservoir simulator, highlighting potential problems and pitfalls, and demonstrating its performance across dif ferent computing platforms for real-field models.