Large-scale computational fluid dynamics (CFD) simulations—used in refining, petrochemicals, power generation, renewable fuels, and more—are crucial to heavy industries to make decisions in the design, scale-up, optimization, and troubleshooting of their processes. However, multiphase simulation of industrial scale fluid-particle scenarios with millions or billions of particles has traditionally been computationally intensive, and only performed occasionally.
Advancing CFD Simulations with NVIDIA Accelerated Computing
CPFD Software’s Barracuda Virtual Reactor is a physics-based engineering software package used to simulate the three-dimensional, transient, fluid-particle hydrodynamics, heat transfer, and chemical reactions in industrial reactors and other fluid-particle systems. To drive further acceleration, CPFD benchmarked Barracuda Virtual Reactor on the NVIDIA GH200 Grace Hopper Superchip, provided in collaboration with Mark III Systems.
We’re excited to share the initial results, which show the Barracuda Virtual Reactor achieving up to 2x faster performance on the GH200 Superchip compared with NVIDIA’s previous architecture generation.
Key CFD Benchmarks Highlight Breakthrough Performance
The graph illustrates the dramatic improvement across a set of benchmarks, which include four models of various industrial-scale reactors with millions of computational particles, hundreds of thousands of cells, and detailed chemical reactions. Numbers indicate the calculation time achieved in a day for a given case.
Simulation speeds for Barracuda Virtual Reactor on NVIDIA’s AI platform. Barracuda Virtual Reactor achieves up to 2x faster simulation speed using the NVIDIA GH200 Grace Hopper Superchip.
- Case 1 – Fluidized bed reactor for chemical production of acrylonitrile.
The GH200 Superchip offered a 1.85x speedup over the previous architecture - Case 2 – Hydrodynamic study of a circulating fluidized bed for energy production (isothermal and non-reacting).
The GH200 Superchip offered a 49x speedup over the previous architecture generation. - Case 3 – Same system as Case 2 with increased particle and cell resolution and with chemistry and heat transfer enabled.
The GH200 Superchip offered a95x speedup over the previous architecture generation. - Case 4 – Fluidized bed reactor for petrochemical processing.
The GH200 Superchip offered a 1.75x speedup over the previous architecture
The NVIDIA GH200 Superchip sets a new standard for simulation speed, nearly doubling the performance of our previous best results achieved with NVIDIA A100 GPUs.
Further benchmarking showed Barracuda Virtual Reactor is 43% faster on the GH200 Superchip than on an integrated x86 CPU and NVIDIA H100 GPU.
These speedups demonstrated the power of NVIDIA accelerated computing and the assiduous work performed by CPFD Software’s R&D engineers, who have leveraged the NVIDIA® CUDA® software platform, integrating key accelerated libraries within Barracuda Virtual Reactor’s numerical solver.
Get Started with Barracuda Virtual Reactor
These performance gains represent a major leap forward for our users; faster CFD simulations open doors to tackling larger and more complex industrial reactor models and allow for virtual testing of a broader space of operational parameters.
If you’re curious about how Barracuda Virtual Reactor paired with the NVIDIA GH200 Superchip can transform your CFD workflows:
- Contact us to get started
- Reach out via email
- Send us a message on LinkedIn
Note: CPFD was an early adopter of NVIDIA’s accelerated computing platform in CFD, and that work continues. Our recent tests of the NVIDIA GH200 Grace Hopper Superchip showed great speedups. This is an area of continued research for CPFD, and we look forward to having this capability supported in future releases.