Introducing the 4A’s of Next-Gen Multiphysics CFD Solution

Many transformational challenges faced today across industries such as transportation, environment, health, defense, and space systems are inextricably linked to a deep understanding of fluid mechanics. For instance, the acoustic noise generated around automobile rearview mirrors, resulting from recirculating flow-inducing pressure oscillations on side windows, can significantly impact cabin noise levels. Computational fluid dynamics (CFD) codes allow for the precise prediction and analysis of such phenomena, taking into account the intricacies of the mirror's design and airflow behavior while requiring far fewer resources compared to experimental approaches.

The current market requires CFD solutions that provide accuracy, automation, speed, and integration with artificial intelligence (AI)—principles that underpin Cadence Fidelity CFD. This blog post will shed light on how these four pillars, as stated by Frank Ham (VP R&D) in his presentation at CadenceLIVE Silicon Valley 2024, collectively referred to as the 4A's, position Cadence Fidelity CFD software as the next-generation multiphysics CFD solution.

A Quick Glimpse at Fidelity CFD

Cadence Fidelity CFD is not merely a suite of solvers but is an extensive ecosystem that has evolved through strategic acquisitions and organic developments over the past five years. This ecosystem encompasses a comprehensive range of tools and technologies designed for model building, solving, and learning within the realm of CFD and Multiphysics CFD. The driving principles behind all the products under Fidelity CFD are Accuracy, Automation, Acceleration, and AI, explained briefly below with a few examples.

Products under Fidelity CFD

Accuracy

One of the perennial challenges engineers face is reliance on physical testing to validate and certify products designed using CFD. Despite the advancements in simulation technology, physical testing remains indispensable for achieving the exceptional accuracy required for final product validation. Take, for instance, the aerospace industry, where aircraft designs must undergo rigorous physical testing to meet stringent safety and performance standards. However, emerging advanced methodologies such as large eddy simulation (LES) offer promising alternatives. While computationally demanding, LES provides a comprehensive and highly detailed representation of fluid flow phenomena. This method can bridge the gap between simulation and physical testing by delivering near-experimental levels of accuracy, thereby reducing the reliance on extensive physical tests and accelerating the design and certification process.

Comparison of sound power level (PWL) from experiment (black symbols) and LES (red line) for the SDT fan with low-noise OGV at approach condition (Brès et al. 2023)

Automation

Automation is indispensable in CFD, particularly within the Cadence Fidelity suite, where it plays a critical role. Automation is woven throughout the CFD workflow using Python-based scripting, ensuring consistency and superior control over simulations with minimal manual intervention. This leads to significant efficiency gains, especially in repetitive tasks.

Take automotive preprocessing as an example. Automotive designs involve highly complex geometries with potentially hundreds of thousands of parts from the CAD system, often missing elements. Fidelity products streamline this workflow through extensive automation. The CAD import process efficiently filters out interior cabin elements, automatically detects overlaps, selects and deletes duplicate objects, identifies wet surfaces, and creates sealing surfaces. For instance, the AutoSeal feature allows users to designate "wet" (exterior) and "dry" (interior) points, automating the gap and joint filling processes, thus expediting CFD simulations.

AutoSeal technology detects "wet" and "dry" points and automatically seals gaps with surfaces in minutes

Voronoi-based grid generation is another standout feature in Fidelity CFD, leveraging a high level of automation. This technique creates a top-notch mesh around intricate geometries, ensuring uniformity and improving simulation accuracy and convergence rates. Traditional mesh generation methods often result in high-quality meshes near the surface, but as layers interact, they create lower-quality meshes, impacting simulation convergence rates and accuracy. Voronoi-based grid generation provides a more consistent and effective solution, enhancing the overall simulation process.

Comparison of Voronoi-diagram mesh with typical RANS mesh for a bike rider model

Acceleration

Acceleration is a critical component of computational science, particularly in CFD. In 2023, NVIDIA’s CEO Jensen Huang’s keynote presentation at a GTC conference demonstrated the substantial acceleration in a simulation on switching from a CPU to a GPU compute environment. Using Amazon cloud pricing for comparison, a graph presented by Jensen revealed that GPU instances offer over nine times greater throughput than CPU instances for the same cost. This staggering figure not only underscores performance improvement but also signifies cost savings—nine times cheaper for the same throughput.

Comparison between CPU and GPU on throughput for the same cost, and cost and energy savings for the same throughput

Additionally, for equivalent throughput, energy consumption is far lower for GPUs compared to CPUs. This efficiency stems from CFD algorithms being more suited to GPUs' parallel architecture. The key measure here is memory bandwidth, with high-bandwidth memory (HBM) playing a crucial role. Effective latency hiding in a well-designed solver allows more efficient data transfer from memory to compute units, leading to performance scaling with memory bandwidth.

Historical data shows that GPUs, even early models, offered significantly higher memory bandwidths. Current-gen GPUs like the H100, A100, or MI200 pack the equivalent computational power of approximately 1000 CPU cores—a capability that once required a sizeable cluster is now achievable with a single GPU.

Cadence capitalized on this transformative potential with Millennium M1, an accelerated CFD supercomputer available both in the cloud and on-premise. Millennium M1 allows enterprises to migrate their CFD workloads to GPU environments, achieving unprecedented throughput increases. It is both secure and accessible and supports high-fidelity simulations that previously required several weeks to complete. The numbers below highlight the revolutionary impact of Cadence’s HW/SW integration, drastically reducing the time required for high-fidelity simulations.

Runtime using two-millennium nodes takes only half the hours typically reported for such simulations

Artificial Intelligence

Artificial Intelligence (AI) is set to transform CFD through data-driven techniques that enable predictive models, and this domain is referred to as deep learning. Cadence is pioneering the implementation of geometric deep learning. Unlike traditional methods, geometric deep learning uses the shape of objects—such as turbo machines or automotive vehicles—as inputs and leverages simulation data to predict integral quantities like drag performance or efficiency losses.

AI used to relate automotive drag prediction to vehicle shape sampled in various ways

An important aspect of this technology is the volume of data required to effectively train AI models and their sensitivity to shape variations impacting performance. Millennium M1 excels in rapidly generating precise data, facilitating robust AI model training. For example, after conducting over 100 automotive drag simulations involving the transformation of a car's shape, the AI model learned to predict drag with significant accuracy by analyzing the vehicle's shape through visual cues rather than traditional parameterization. This emphasizes the promising role that AI will play in advancing CFD simulations.

Cadence Fidelity CFD, with its foundational 4A’s—Accuracy, Automation, Acceleration, and Artificial Intelligence—addresses simulation challenges head-on, providing a next-generation Multiphysics CFD solution that meets the demands of modern industry. By leveraging LES for high-fidelity accuracy, implementing streamlined workflows through automation, harnessing the computational power of GPUs for accelerated performance, and pioneering geometric deep learning for predictive modeling, Cadence Fidelity CFD sets a new standard in the CFD landscape. As industries continue to evolve and the complexity of engineering problems grows, integrating these advanced methodologies will be essential in driving innovation, efficiency, and sustainability across various sectors.

References

Bres, Guillaume A., Wang, K., Emory, M., et al., “GPU-accelerated large-eddy simulations of the NASA fan noise source diagnostic test benchmark,” AIAA 2023-4299. June 2023, https://doi.org/10.2514/6.2023-4299

To watch the on-demand video of Frank Ham’s keynote Cadence Fidelity CFD: Accuracy, Acceleration, Automation, and AI at CadenceLIVE Silicon Valley 2024, click the button below.