Cadence, NVIDIA, and Solar Turbines Collaborate on AI Physics

Accelerated computing and advanced simulation technologies are changing the game for the traditionally experiment-heavy power generation industry by offering unparalleled efficiency and precision. Full-scale industrial models with complex design features and multiple components can now be analyzed in extraordinary detail through GPU-accelerated simulations. This enables an in-depth study of flow physics, flame-flow interactions, and the evaluation of multiple design configurations to achieve optimal solutions. Solar Turbines, a subsidiary of Caterpillar Inc., has collaborated with Cadence and is leveraging Cadence technologies powered by NVIDIA to conduct scalable, high-fidelity reacting flow simulations for gas turbines. By utilizing GPU-accelerated Cadence Fidelity CharLES  on NVIDIA Blackwell GPUs, the team can now run reacting flow simulations with over 1 billion cell grids in less than a day.

Why is Scalable High-Fidelity Simulation Essential for Large-Scale Industrial Models?

In recent years, there has been a shift toward using computational fluid dynamics (CFD) tools for cost-effective optimization of industrial power generation systems under diverse flow conditions. However, the application of CFD in the industry is often restricted to simulating individual system components in isolation. These components are then manufactured and assembled, only to find that the performance of the full system often falls short of expectations. This inefficiency increases both time-to-market and costs.

While the widely used Reynolds-averaged Navier-Stokes (RANS) approach is a staple in the field, its intrinsic limitations in capturing flow and flame unsteadiness hinder its ability to handle complex combustion phenomena. In contrast, large eddy simulation (LES) is highly regarded for its ability to accurately capture transient, large-scale turbulent structures, making it better suited for predicting combustion instabilities. However, LES for full-scale industrial models remains computationally challenging due to the need to resolve intricate flow and flame features, complex geometries, and extensive computational domains.

These limitations highlight the need for scalable, GPU-accelerated LES simulation software that can effectively handle the demands of simulating large-scale models. Such technologies enable engineers to fully understand the physics at play, and more importantly, to capture how interactions between system components influence overall performance. By leveraging advanced simulation tools that can handle system-wide complexities, the industry can achieve better-informed decision-making, reduced design cycle times, and more reliable outcomes.

Solar Turbines, Simulates Large-Scale Reacting Flow Simulations with Cadence tools powered by NVIDIA

The collaboration between Cadence and Solar Turbines, leveraging NVIDIA technologies, is a groundbreaking example of leveraging advanced computational technologies to tackle the complexities of large-scale reacting flow simulations.

Exhaust Gas Recirculation Studies in Collaboration with Argonne National Laboratory

 High-fidelity simulations of Solar Turbines’ Taurus 60 SoLoNOx combustor were conducted in collaboration with Argonne National Laboratory to investigate the effects of exhaust gas recirculation (EGR) on combustion instabilities and emissions (Kabil et al., 2025). With Cadence’s Fidelity CharLES Solver and a flamelet progress variable (FPV) turbulent combustion model, the simulations achieved excellent accuracy by capturing detailed turbulent flow structures and combustion chemistry. These efforts were validated against experimental data, offering insights into the impact of EGR under varying conditions while ensuring scalability and efficiency.

Geometry of the Taurus 60 gas turbine combustor. (a) Full annular combustor (b) Double injector sector (Kabil et al., 2025)

Breakthrough Simulation Performance with NVIDIA Accelerated Computing

NVIDIA AI infrastructure played a critical role in the success of these simulations. The initial simulations were performed on NVIDIA GPU clusters at Argonne National Laboratory, with scalability tests demonstrating approximately 90% efficiency on 80 GPUs for a case involving roughly one million cells per GPU. This translated to a computational speedup of nearly two orders of magnitude compared to traditional CPU-based methods.

Building on this progress, advanced simulations of the large-scale model utilizing finer meshes were conducted on NVIDIA GB200 NVL72. These scalable, high-fidelity simulations achieved an impressive 94% parallel efficiency on 128 GPUs and sustained 75% efficiency when scaled up to 512 GPUs. Remarkably, this performance enables simulations of a full-360º gas turbine combustor, comprising over 1 billion cells, in just 14 hours, using 512 GPUs. To match this level of performance with CPUs alone would require the computational power of an entire data center.

Another example that demonstrates the capabilities of NVIDIA AI infrastructure is the ability to generate high-fidelity simulation data for airframe simulations. Utilizing the power of the NVIDIA GB200 platform, accessible through the Cadence Millennium M2000 AI Supercomputer, thousands of simulations can be efficiently generated within just a few weeks, setting a new benchmark in computational performance and precision.

These case studies, focusing on large-scale reacting flow simulations and the generation of high-fidelity airframe simulation data, showcase how advanced CFD solutions and accelerated compute technology propel industrial modeling and AI-driven physics. By enabling rapid design iterations, improved efficiency, and optimized performance, these advancements are paving the way for next-generation systems. Cadence, with NVIDIA, is innovating and setting a new benchmark in computational engineering, redefining the future of technological development.

To learn more about the collaboration between Cadence and NVIDIA for simulating large-scale industrial models, read the NVIDIA blog.

Reference

Kabil, I, Xu, C, Shunn, L, Wang, J, Sung, Y, Johnson, D, & Steele, C. "GPU Accelerated Large Eddy Simulations of Combustion Instabilities in Industrial Gas Turbines Under EGR Conditions." Proceedings of the ASME Turbo Expo 2025: Turbomachinery Technical Conference and Exposition. Memphis, Tennessee, USA. June 16–20, 2025. V03AT04A002. ASME. https://doi.org/10.1115/GT2025-151173