Cadence and Toyota Advance Automotive Simulation on NVIDIA AI Infrastructure

Cadence is collaborating with Toyota Motor Europe (TME) to evaluate high-fidelity external aerodynamics and aeroacoustic simulation on the TME Aygo X model using NVIDIA Blackwell GPUs. The work demonstrates how accelerated computing can reduce turnaround time for computational fluid dynamics (CFD) workflows while supporting the generation of richer simulation datasets for building the foundation for physics AI models.

The benchmarking effort uses the RTX PRO 6000 Blackwell Server Edition and NVIDIA DGXB200 to run Fidelity Charles Solver, the large eddy simulation (LES) solver in Cadence Fidelity CFD Software, across a mesh configuration from a 2024 Aygo X benchmarking study. The goal is to assess scalability, solution time, and the potential to shift more aerodynamic and noise-suppression validation from physical testing to simulation-led development.

Benchmarking the Aygo X on Blackwell GPUs

This study evaluates a full-vehicle body refinement strategy using a mesh configuration based on clipped Voronoi diagrams designed to resolve external aerodynamic flow features. The configuration applies uniform refinement across the vehicle's complete external body and contains 178 million control volumes (MCVs).

The mesh configuration, along with the low dissipation of Fidelity Charles operators, enables simulation of fine-scale unsteady flow structures that influence external aerodynamics and aeroacoustic analysis. This approach enables a more detailed virtual wind tunnel analysis, offering accuracy, efficiency, and scalability in production-scale automotive CFD workloads.

"Our collaboration with Cadence demonstrates how scalable high-fidelity CFD on NVIDIA GPUs can achieve turnaround time significantly lower than our CPU-based workflow, without sacrificing accuracy," said Antoine Delacroix, senior manager, Vehicle Performance Engineering R&D at Toyota Motor Europe. Such fast aerodynamics simulation-based workflow supports our journey to make ever better cars."

GPU Scalability Reduces Simulation Turnaround Time

Preliminary results show strong scalability for high-fidelity aerodynamic simulation on Blackwell GPUs. For the Aygo X model, the graphs below show the simulation speedup and the decreasing wall time per second of simulation time with increasing GPU count.

Simulation results of the Aygo X external aerodynamics: (i) simulation speed up vs. GPU count on RTX Pro 6000 Blackwell Server Edition (left), (ii) wall time per second of simulation time vs. GPU count on RTX Pro 6000 Blackwell Server Edition and DGX B200 (right)

For the same Aygo X simulation, the right-hand graph shows that the eight-GPU B200 configuration reaches convergence about 5X faster than the RTX PRO 6000 Blackwell Server Edition. It also shows the difference in per-GPU performance: the RTX PRO 6000 requires 32 GPUs to approach the convergence time achieved by the B200 with only eight GPUs.

This study represents the second major stage of GPU acceleration for the Fidelity Charles Solver, progressing from the first-generation Millennium platform to the NVIDIA Blackwell-powered Millennium M2000 Supercomputer. For comparable large-scale simulations, internal engineering estimates indicate that matching the demonstrated GPU throughput with CPU-only infrastructure could require thousands of CPU cores, potentially in the range of approximately 1,500 CPU cores for the RTX PRO 6000 configuration and approximately 5,000 CPU cores for the DGX B200 configuration.

These figures are directional estimates based on a separate aircraft simulation, not measured results from the Aygo X benchmark, and should therefore be presented as context rather than a direct performance claim. The broader point is that production customers have moved these extreme-scale Fidelity Charles workloads to GPUs because the required CPU infrastructure is no longer practical at this level of simulation.

"High-fidelity simulation is foundational to how automakers design, test, and refine the next generation of vehicles," said Tim Costa, vice president and general manager of industrial and computational engineering at NVIDIA. "By accelerating the Fidelity Charles Solver's large eddy simulations with NVIDIA CUDA-X, Cadence is helping engineers evaluate aerodynamic and acoustic tradeoffs across a broader design space and refine their designs before physical testing."

The results highlight the value of higher per-GPU performance for production-scale CFD. While the RTX PRO 6000 Blackwell Server Edition shows improved turnaround time as GPU count increases, the DGX B200 system achieves comparable convergence time with significantly fewer GPUs. This suggests that next-generation GPU architectures can help engineering teams reduce infrastructure requirements while increasing simulation throughput. For automotive workflows, this can translate into more design iterations, broader parametric studies, and larger high-fidelity datasets for future AI physics development.

An Integrated Cadence Workflow from Pre-Processing to AI

The workflow brings together Cadence BETA CAE Simulation Solutions and Cadence Fidelity CFD Software in a connected simulation workflow. Cadence BETA CAE ANSA Pre-Processor is used for pre-processing, the Fidelity Charles Solver performs the high-fidelity flow simulation on NVIDIA GPUs, and Cadence BETA CAE META Post-Processor supports post-processing, visualization, and reporting. For future AI physics applications, Cadence BETA CAE KOMVOS SDM Console can manage simulation data and help develop predictive models from high-volume CFD results.

Building the Foundation for AI Physics

Beyond turnaround time, GPU-accelerated simulation creates an opportunity to generate high-fidelity datasets at a scale that is difficult to achieve with CPU-only infrastructure. The Millennium M2000 Supercomputer greatly simplifies the adoption of accelerated CFD simulation. Running the Fidelity Charles Solver at scale with Millennium technology makes broad parametric exploration more practical, giving engineering teams the volume and diversity of data needed to train AI physics models that reflect the complexity of real-world automotive design.

With the KOMVOS SDM Console, high-volume simulation results can be retained as compact surrogates, making it easier to retrieve prior simulation knowledge, compare design variants, and apply cross-project learning. The approach aligns with a broader industry shift toward large physics models, where pre-trained models can be fine-tuned on targeted datasets to reduce the amount of new simulation data required for accurate prediction.

"Our expanding multiphysics portfolio, coupled with NVIDIA GPU acceleration and our collaboration with Toyota, is enabling a new scale of engineering simulation for vehicle development," said Joe Cole, corporate vice president, EMEA at Cadence. By bringing high-fidelity physics, accelerated computing, and broader design-space exploration together, we are helping engineers evaluate more configurations and operating conditions in far less time and ultimately enabling faster convergence toward optimized vehicle designs."

For automotive development teams, this creates a path from high-fidelity CFD to faster engineering insight: generate accurate simulation data with Fidelity Charles Solver, review and automate results in META Post-Processor, manage and learn from simulation assets in KOMVOS SDM Console, and use AI models to accelerate future design exploration.

Explore Physical System Simulation Solutions

Discover our range of solutions for connected, system-level engineering, and talk to an expert to learn how they can support your design challenges.