GPU Speed-Up for the Implicit Navier-Stokes Solver

In this study, an in-house higher-order, high resolution implicit CFD solver for Favre-averaged Navier-Stokes equation with two-equation turbulence model, which including fourth-order MUSCL TVD scheme for Right-Hand-Side and LU-SGS-GE scheme for Left-Hand-Side, is implemented to run on a single Graphic Process Unit (GPU) using the Compute Unified Device Architecture (CUDA).The partly parallel model for implicit scheme is introduced and we have got a good acceleration using this partly parallel model. The case of VKI 59 is used to test the accelerations. Results obtained on a single GPU (NVDIA GeForce 560Ti, Memory: DDR5 1G 2004 Mhz) are compared with the results obtained on the single CPU (Intel i5 2.8–3.0 GHz, Memory: DDR3 8G 800 Mhz). The speed-up is about 25. The fourth-order MUSCL TVD used in the code is extended to the simulation of multi-species, and also speed-up by using GPU. The case of a single stage turbine with hot-spot inlet boundary is used to test the code with multi-species scheme using GPU. Furthermore, in the study, the shared memory is used to speed-up the code, which is introduced in details in the paper. In the end, the influence factors of the acceleration are discussed.Copyright © 2013 by ASME