Automotive

Boosting Vehicle Aerodynamics with NVIDIA GPUs

Objective

GAC R&D Center deployed NVIDIA GPUs on its hybrid cloud platform to achieve record low drag coefficient on a new concept car design.

Customer

Guangzhou Automobile Group Co., Ltd. (GAC Group)

Partner

Altair ultraFluidX

Use Case

Computational Fluid Dynamics (CFD) Simulation

Technology

NVIDIA V100 SXM2 GPUs

Achieving Global Competitiveness

GAC R&D Center (GAC RDC), established in 2006, is a wholly owned subsidiary of Guangzhou Automobile Group Co., Ltd. (GAC). It’s GAC Group’s technology management department and R&D system hub for new product development for its two brands, GAC Trumpchi and GAC NE, as well as new technology roadmaps and major R&D implementations.

To build a globally competitive brand and demonstrate its technological capabilities and future-proof design ability, GAC proposed a new concept car at Auto Guangzhou–responsible for the R&D work of this vision–and set an important objective to create a new record for lowest drag coefficient.

Exploring Efficient and Accurate CFD Simulation Technologies for Design Optimization

Most of the designs required evaluation and optimization through computational fluid dynamics (CFD) simulation. A key challenge was to decide which CFD simulation technology to use to improve overall simulation accuracy and efficiency.

Conventional CFD approaches were used to enable acceleration and efficiency with the Message Passing Interface (MPI) mechanism for multi-core CPU, multi-thread parallel computing and high-performance computing (HPC) resource scheduling. This not only created a demanding requirement for CPU cores in the HPC clusters; it also presented challenges such as high energy consumption and high maintenance cost to use the clusters.

CFD software based on conventional approaches often demands higher grid quality, complex pre-processing, and heavy investment in manual processes, which leads to difficult autonomous implementation. To obtain high-precision simulation results, larger-scale grids and transient simulation are often required, which inevitably leads to a sharp increase in computing resource consumption.

Image courtesy of GAC R&D Center

GAC R&D Center is Guangzhou Automobile Group’s technology management department and R&D hub.

GAC R&D Center built a transient simulation of automobile outflow field that resulted in a new record for lowest drag coefficient.

Image courtesy of GAC R&D Center

GAC RDC Achieves Record-Breaking Drag Coefficient with GPU-Powered HPC Platform

To meet the project requirements for R&D, GAC RDC deployed NVIDIA V100 SXM2 Tensor Core GPUs on its heterogeneous hybrid cloud platform for high performance computing, each featuring 5,120 CUDA® cores. With a double precision floating-point computing capability of 7.8 teraFLOPS (TFLOPS), its GPU parallel computing efficiency dramatically improved compared to CPUs on the same model with the same simulation accuracy. A single project has about 120 million CFD grids (particles), and the simulation calculation takes about 10 hours.

The GAC RDC aerodynamics team adopted the Altair ultraFluidX CFD software based on GPU double-precision computing technology with NVIDIA V100 computing resources. In less than six months, the team completed over 200 transient CFD simulations of the vehicle outflow field, resulting in several viable solutions. The simulation value of drag coefficient in the demo state was 0.147 and the test value was 0.146 (per the Shanghai Automotive Wind Tunnel Center of Tongji University), setting a new record on lowest drag coefficient with impressive results over the previous record of 0.19. Compared to the transient CFD simulation based on conventional approaches, the manual effort required for modeling was reduced by nearly 60 percent and the total simulation time was shortened by about 70 percent.

Powered by the hybrid cloud platform for HPC heterogeneous computing, GAC RDC has built an agile system for vehicle aerodynamics development that combines holographic CFD simulation with wind tunnel tests— effectively improving its development efficiency and accuracy. This system helps to ensure its leadership among domestic OEMs in large-scale CFD collaborative simulation and ultra-low vehicle drag coefficient. For example, the drag coefficient of the recently launched GAC Trumpchi GS4 Coupe is only 0.295, which is far lower than that of comparative models in its vehicle segment.

Image courtesy of GAC R&D Center

Why NVIDIA

Accelerate vehicle aerodynamics development with lower modeling inputs and faster time to simulation

Automate high-precision, transient CFD simulations with GPU parallel computing

Reduce the high energy consumption and maintenance costs of using multi-core CPUs

Simulation-Driven Design for Low Drag Coefficient in Battery-Powered Electrical Vehicles

With strict national regulations on fuel consumption and higher requirements for battery-powered electrical vehicle (BEV) range, the development of vehicle aerodynamics is becoming more important than ever. As wind drag accounts for a major part of driving resistance at high speed, reduction of drag coefficient is one of the major strategies for OEMs to reduce energy consumption and emission. Due to the very high cost of wind tunnel tests and the general lack of owned full-scale wind tunnel test chambers by domestic OEMs, the cost-efficient, integrated CFD simulation solution will play a critical role in the design of simulation-driven models that feature an ultra-low drag coefficient.