Investigate the Computational Performance of Many-Core Graphics Processing Unit and Coprocessor for Fluid-Film Lubrication Analysis

An efficient numerical computation for dealing with fluid-film lubrication study is an important analysis tool. In many cases the computation requires to obtain the solutions of various Reynolds equations and some other related equations, such as the equations considering the compressibility of the lubricant, the variation of viscosity, and the deformation of the bearing surfaces. As a result, the governing equations, such as in thermohydrodynamic lubrication, may include generalized Reynolds equation, energy equation, and elastic deformation equation. In recent years, the advancement of high performance computing is developed in a fast pace in many areas. These include multi-core CPUs/coprocessors and graphics processing units (GPUs), as well as the open architecture of parallel programming paradigms, mainly OpenMP and OpenACC. The OpenACC standard is for programming in GPUs, while the OpenMP is the main standard for processors with multiple processor cores. In this two-year project, the high-performance Nvidia GPUs are to be incorporated with the new version of OpenACC parallel programming language. The computing efficiency of the newly developed coprocessors from Intel is also to be investigated using the OpenMP coding. The efficiency of the both approaches are to be assessed by using the numerical model for the transient thermohydrodynamic lubrication analysis. The main topics in the first year of the project are: (1) Apply the OpenACC (with GPUs) to solve the transient lubrication model to investigate the performance of the GPU computing as well as to develop methods to increase the parallel efficiency; (2) Evaluate the effectiveness of solving transient problem using efficient parallel computing algorithm to obtain the steady-state solution. The main topics in the second year of the project are: (1) Apply the OpenMP (with Intel coprocessor) to solve the transient lubrication model to investigate the performance of the multithreaded computing as well as to develop methods to increase the parallel efficiency; (2) Combine the computing power of the GPU and coprocessor to further increase the performance of the numerical computation for fluid-film lubrication analysis.

Project ID:PB10408-5730
External Project ID:MOST104-2221-E182-065