Abstract
A high fidelity flow simulation for complex geometries for high Reynolds number (Re) flow is still very challenging, requiring a more powerful HPC system. However, the development of HPC with traditional CPU architecture suffers bottlenecks due to its high power consumption and technical difficulties. Heterogeneous architecture computation is raised to be a promising solution to the challenges of HPC development. GPU accelerating technology has been utilized in low order scheme CFD solvers on the structured grid and high order scheme solvers on unstructured meshes. The high-order finite difference methods on structured grids possess many advantages, e.g., high efficiency, robustness, and low storage. However, the strong dependence among points for a high-order finite difference scheme still limits its application on the GPU platform. In the present work, we propose a set of hardware-aware technology to optimize data transfer efficiency between CPU and GPU, as well as communication efficiency among GPUs. An in-house multi-block structured CFD solver with high order finite difference methods on curvilinear coordinates is ported onto the GPU platform and obtains satisfying performance with a speedup maximum of around 2000x over a single CPU core. This work provides an efficient solution to apply GPU computing in CFD simulation with specific high order finite difference methods on current GPU heterogeneous computers. The test shows that significant accelerating effects can be achieved for different GPUs.
Highlights
Computational fluid dynamics (CFD) is one of the most important research methods in fluid mechanics, which highly relies on the computational capability of the computer, especially for accurately simulating realistic engineering flows
The boom of graphics processing unit (GPU) computing in the past several years shows the power and potential of GPU on computing, and it attracts more and more researchers to exploit its application in their fields, including CFD simulation
We analyze the characteristics of architectures of current GPU servers and propose a set of techniques to improve the efficiency of data transfer between CPU and GPU and efficiency of communication between GPUs
Summary
Computational fluid dynamics (CFD) is one of the most important research methods in fluid mechanics, which highly relies on the computational capability of the computer, especially for accurately simulating realistic engineering flows. Et al [7, 8] develop a multiple GPU algorithm in hypersonic flow computations with the second-order finite volume method on curvilinear coordinates, and obtain 20x to 40x speedup accelerating with an Nvidia GTX1070 GPU compared to an eight cores Intel Xeon E2670 CPU. For a complicated coordinate system and high order schemes, the flow solvers get an inferior accelerating effect, which is far from the computational requirement for higher fidelity and larger-scale simulation. It is easy to improve precision, the numerical stability of this type of method in solving high-speed flows is poor, which limits its application This method requires a large amount of storage, limiting its application in large-scale simulations in GPU computing. The CPU of the server is Intel Xeon E5 2680 v3, with 12 cores
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
