SimScale: Unparalleled CFD Speeds with Parallel Computing
0 likes122 views
SimScale offers a cloud-based engineering simulation platform that optimizes vehicle aerodynamics using computational fluid dynamics (CFD) with unprecedented speed and cost-efficiency. Leveraging AWS's high-performance computing and NVIDIA GPUs, the platform reduces simulation times from weeks to hours while maintaining high accuracy in drag coefficient calculations. The integration of a robust solver and advanced cloud technology results in significant improvements in automotive design processes.
![Less than 5% deviation for Drag Coefficient for
all models
~2.5% deviation for Fastback and Notchback
models
Drag Coefficient results accurately represent
changes in geometry and trends
5% and under deviation considered very
accurate for industry
Note: Alternative experiments (John et.al [4]) on the
Estateback model indicate a drag coefficient of 0.309,
which would give a deviation of 1.6%
Fastback Estateback Notchback
Experimental 0.275 0.319 0.277
Simulation 0.268 0.304 0.271
% Error 2.55% 4.70% 2.24%
Results](https://image.slidesharecdn.com/simscaleunparalleledcfdspeedswithparallelcomputing-201119085735/85/SimScale-Unparalleled-CFD-Speeds-with-Parallel-Computing-12-320.jpg)
![SimScale / Pacefish®
Experimental Results (Heft et.al [5][6])](https://image.slidesharecdn.com/simscaleunparalleledcfdspeedswithparallelcomputing-201119085735/85/SimScale-Unparalleled-CFD-Speeds-with-Parallel-Computing-15-320.jpg)
![SimScale / Pacefish®
Experimental results (Heft et.al [5][6])](https://image.slidesharecdn.com/simscaleunparalleledcfdspeedswithparallelcomputing-201119085735/85/SimScale-Unparalleled-CFD-Speeds-with-Parallel-Computing-17-320.jpg)
SimScale: Unparalleled CFD Speeds with Parallel Computing
- 1. © 2020, Amazon Web Services, Inc. or its affiliates. All rights reserved. SimScale: Unparalleled CFD Speeds with Parallel Computing Track: AWS HPC Workloads
- 2. Maximise Fuel Economy by optimizing vehicle shape using computational fluid dynamics CAD Assembly File CFD Cleaned CAD Meshed Model Results ~10 days 7 days Hardware Dependent What if you had results by lunch direct from the CAD Assembly? Automotive Aerodynamics
- 3. What are we talking about? Idea Implementation Impact Reduce automotive aerodynamics workflows from weeks to hours Weave the SimScale platform with: ● An efficient, robust and accurate solver ● High Performance cloud computing Turnaround times unseen by the automotive industry, and an enhanced, cost-efficient iterative design process
- 4. SimScale Platform We created the world’s first cloud-based engineering simulation platform. ● Fluid dynamics (CFD) ● Solid mechanics (FEA) ● Thermodynamics All accessible via a web browser.
- 5. Pacefish Solver ● Isothermal Transient Flow Solver based on the Lattice Boltzmann method ● Developed by Numeric Systems GmbH ● Fully integrated into SimScale ● Multi-GPU enables extremely fast turnaround times ● Runs on AWS P3 instances with latest generation Nvidia GPUs ● Turbulence models: Smagorinsky (LES), SST-DDES, SST-IDDES (uRANS-LES) and k-omega SST (uRANS) Dedicated solver enabling accurate, fast and robust CFD
- 6. Estateback E_D_wM_wW Fastback F_D_wM_wW Notchback N_D_wM_wW DrivAer The industry standard validation cases. Simultaneously. In just a few hours.
- 7. It takes 3 things operating in unison to achieve such a task. An intuitive, easy and robust platform, followed by a powerful, fast and accurate solver, and finally a lot of computing power! What are we talking about?
- 8. Lattice Boltzmann Method The lattice boltzmann method implemented in Pacefish® by our partners at Numeric Systems is a robust, accurate, GPU accelerated solver D3Q19 Lattice Cell Particle Methods LBM Navier Stokes Microscopic Mesoscopic Macroscopic
- 9. The solution is broken down into thousands of domains and processed on cloud GPUs at our partner AWS Nvidia V100 in AWS P3 instance AWS Data Centre Ireland High Performance Computing
- 10. AWS was first in the cloud to offer NVIDIA V100 Tensor Core GPUs via Amazon EC2 P3 instances. To increase performance and lower cost, AWS has announced the new EC2 instances based on the NVIDIA A100 Tensor Core GPUs. AWS expect EC2 instances based on NVIDIA A100 GPUs to set new performance benchmarks. For more information about EC2 instances based on NVIDIA A100 GPUs and potentially participate in early access, see here: https://pages.awscloud.com/ec2_instances_nvidia_a100.html Nvidia GPUs on AWS
- 11. ● Greater than 95% accuracy for Drag Coefficient on all models ● Accurate Pressure Coefficient correlation ● 162 million cells ● 10 minute setup time ● Fully transient ● All in the Cloud ● Solved in less than 8 hours Results Upper left: Isovolumes of above and below 17.5 m/s and 12 m/s velocity Lower left: Isovolumes of above and below 30 Pa and 10 Pa (Gauge) 180M cell mesh
- 12. Less than 5% deviation for Drag Coefficient for all models ~2.5% deviation for Fastback and Notchback models Drag Coefficient results accurately represent changes in geometry and trends 5% and under deviation considered very accurate for industry Note: Alternative experiments (John et.al [4]) on the Estateback model indicate a drag coefficient of 0.309, which would give a deviation of 1.6% Fastback Estateback Notchback Experimental 0.275 0.319 0.277 Simulation 0.268 0.304 0.271 % Error 2.55% 4.70% 2.24% Results
- 13. ● State-of-the-art GPUs retail at prices around $10k each ● AWS offers always the latest hardware, on-demand, at competitive price point ● TCO includes power, maintenance, depreciation... ● Pacefish® allows cheap determination of ideal instance size - no over-provisioning! ● AWS P3 instances and SimScale allow aerodynamics CFD computation at unprecedented cost-efficiency Cost Efficiency
- 14. Details: ● 162.2 million cells ● 3mm surface Cells ● Under 9 hours runtime on AWS P3 instance with 8 x Nvidia V100 GPUs Pressure Coefficient Distributions are validated against experimental results published by Heft et al. Pressure Coefficient Fastback
- 15. SimScale / Pacefish® Experimental Results (Heft et.al [5][6])
- 16. Pressure Coefficient Notchback Details: ● 162.2 million cells ● 3mm surface Cells ● Under 8 hours runtime on AWS P3 instance with 8 x Nvidia V100 GPUs Pressure Coefficient Distributions are validated against experimental results published by Heft et al.
- 17. SimScale / Pacefish® Experimental results (Heft et.al [5][6])
- 18. Pressure Coefficient Estateback Details: ● 71.2 million cells ● 4.5mm surface Cells ● Under 4 hours runtime on AWS P3 instance with 4 x Nvidia V100 GPUs Pressure Coefficient Distributions are validated against experimental results published by Heft et al.
- 19. Transient Data
- 20. Conclusion SimScale, Pacefish® and AWS provide a: ● Fully Validated ● Transient ● Web and Cloud based ● More than 10x cheaper ● Lattice Boltzmann Method Solver ● Enhanced by GPU acceleration to reach high levels of accuracy in a fraction of the time. Left: Velocity followed by Vorticity Setup and results publicly available on SimScale
- 21. Thank you! © 2020, Amazon Web Services, Inc. or its affiliates. All rights reserved.