Acceleration of Turbomachinery steady CFD Simulations on GPU

Transcription

1 Acceleration of Turbomachinery steady CFD Simulations on GPU Mohamed H. Aissa 1,2 Dr. Tom Verstraete 1 Prof. Cornelis Vuik 2 1 Turbomachinery Department, Von Karman Institute, Belgium 2 Delft Institute of Applied Mathematics, TU Delft, the Netherlands Unconventional HPC, EuroPar 2016 WS Grenoble,

2 Topic of Interest Topic of Interest: Reduce Fuel Consumption and CO 2 Emission Wikipedia.org

3 Topic of Interest Turbomachinery is about Performance and Efficiency

4 Topic of Interest Axial Jet Engine Source: Wikipedia.org

5 Content Multidisciplinary Optimization CFD simulations on GPU Literature review Implicit RANS Implementation Benchmark Optimization Case

6 Topic of Interest Optimization algorithm Derivative-based optimization fast convergence but.. derivative evaluation could be complicated and problem specific (adjoint, automatic differentiation) Derivative free methods: e.g. Population based Simplicity Black box approach of the evaluation but.. Large number of evaluations min f(x) subject to g x 0

7 Topic of Interest Optimization algorithm Derivative-based optimization fast convergence but.. derivative evaluation could be complicated and problem specific (adjoint, automatic differentiation) Derivative free methods: e.g. Population based Simplicity Black box approach of the evaluation but.. Large number of evaluations min f(x) subject to g x 0

8 CFD: Core of the Optimization CFD much slower than CSM Need for acceleration -> GPU CADO: the VKI in-house optimizer

9 Steady CFD Simulations Simulation with a unique solution for given boundary Conditions. A start solution is advanced iteratively in time until convergence

10 Steady CFD Simulations Simulation with a unique solution for given boundary Conditions. A start solution is advanced iteratively in time until convergence

11 Numerical Scheme: Explicit Time Stepping (β=0): Implicit Time Stepping (β=1): Aissa, M.H., Verstraete, T., Vuik, C. "Aerodynamic Optimization of Supersonic Compressor Cascade using Differential Evolution on GPU". 13th Int. Conf. of Numerical Analysis and Applied Mathematics (ICNAAM 2015)

12 Implicit Time Stepping is more Stable but X 1 X 2 X n

13 Literature Review What to Port only linear solver when it is dominant both assembly and solve is optimal (no communication) Linear solver Library : code maturity but restrictive (petsc-dev, Paralution, AmgX, ViennaCL ) Own code: flexibility Storage format Standard (CSR,DIA ) New (hybrid)

14 CFD Solver (Standard) sa.blogspot. be/2015/10 /forparalutiongpuconversionand.html Implicit Runge-Kutta scheme Xu et Al. JCP 2015

15 CFD Solver (Standard) Implicit Runge-Kutta scheme

16 CFD Solver (On-demand Factorization) Implicit Runge-Kutta scheme

17 CFD Solver (On-demand Factorization) Stop condition relative, absolute or a combination

18 CFD Solver (On-demand Factorization) Stop condition relative, absolute or a combination

19 Benchmark: Flow around LS89 2-Stages Runge-Kutta 1/12

20 9 8 7 Assembly Acceleration Speedups on Coarse Mesh x Assembly speedup Linear solve speedup Global speedup xCores 3xCores 4xCores Standard CPU GPUOn-demand CPU GPU 70% 2xcores 3xcores 4xcores ILU ILU OD Speedups on Fine Mesh 2xCores 3xCores CPU x xCores Standard On-demand GPU 2xcores 3xcores 4xcores ILU ILU OD CPU GPU CPU 30% GPU 10% 90%

21 Linear Solver Acceleration Speedups on Coarse Mesh x 0.7 x 1.2 Assembly speedup Linear solve speedup Global speedup xCores 2xcores 3xcores 4xcores ILU ILU OD CPU GPU Speedups on Fine Mesh 2xCores 3xCores CPU 3xCores 4xCores Standard On-demand x 1.8 x 5.7 4xCores Standard On-demand 2xcores 3xcores 4xcores ILU ILU OD CPU GPU CPU GPU GPU

22 Global Acceleration Speedups on Coarse Mesh x 2.0 x 3.2 Assembly speedup Linear solve speedup Global speedup xCores 2xcores 3xcores 4xcores ILU ILU OD CPU GPU Speedups on Fine Mesh 2xCores 3xCores CPU 4xCores Standard On-demand x 4.8 x 9.6 2xcores 3xcores 4xcores ILU ILU OD 3xCores CPU GPU CPU GPU 4xCores Standard On-demand GPU Suggestion for better Performance assessment are very welcome!

23 Increase of the Speedup for higher Numbers of Runge-Kutta Stages on Fine Mesh Speedup N stages Assembly Solve Global

24 Content Multidisciplinary Optimization CFD simulations on GPU Literature review Implicit RANS Implementation Benchmark Optimization Case

25 Test Topic Case of Interest 3: TU Berlin TurboLab Stator Optimization requirements Objectives: Decrease outfow axial deviation Decrease total pressure loss Considering 3 operating points

26 TurboLab Topic of Interest Manufacturing Constraints N blades = 15 Chord length fixed Casing fixture d=2mm 60 mm d=10mm h=20mm

27 TurboLab: Boundary conditions and summary Inlet P 0 : Pa Inlet T 0 : K 9 kg/s +/- 0.1 Massflow imposed P 2 adapted Objectives: Decrease outfow axial deviation Decrease total pressure loss Considering 3 operating points Inlet whirl angle: 42 Inlet pitch angle: 0

28 Span [-] Parametrization 21 Design variables

29 Turbolab Parameterization

30 Optimization Results 0.17% 1.7 % 60% IT074IND6

31 Optimized Blade

32 Baseline Vs Optimized

33 Baseline Vs Optimized

34 Isentropic Mach Number at mid-span

35 Conclusion Optimization GPU Solver with implicit time stepping On-demand (incomplete) Factorization 10x speedup Aerodynamic shape optimization

36 Future Work Benchmark Case: Transonic Turbine Stator T106c 80 Speedup based on CPU explicit CPU exp GPU imp GPU exp CPU imp 40 GPU Imp GPU Exp. CPU Imp. 0 CPU Exp. 50K 450k 900k Mesh Size

37 Support Hardware Thanks for your attention Mohamed Hassanine Aissa Turbomachinery & Propulsion Department 72, chaussee de Waterloo B Rhode Saint Genese - Belgium aissa@vki.ac.be acknowledgements: