Native GPU support for CFDEM®coupling
CFDEM®coupling 7.1.0 now features native support for running both CFD and coupled CFD-DEM simulations on GPU devices. This is achieved using SPUMA, an advanced framework for open-source CFD running on GPUs. With this integration, users can run their entire coupled simulation—both the Aspherix® particle solver and the fluid solver, directly on the GPU. This eliminates the data transfer bottlenecks often found in hybrid CPU-GPU setups, ensuring maximum computational efficiency.
Performance Benchmarks
The shift to GPU acceleration delivers dramatic reductions in runtime for complex industrial applications. In a benchmark study of a fluidized bed simulation with 1 million particles and 1 million CFD cells, the performance gains were clear:
- CPU (Ryzen 7 3800X): 21.6 hours runtime
- GPU (Nvidia RTX 5070): 7.5 hours runtime
This represents a nearly 3x speedup on a standard desktop GPU compared to an 8-core CPU configuration, allowing for faster design iterations and more detailed sensitivity studies.
Streamlined Ease of Use
We have significantly improved the user environment to reduce setup time and minimize errors.
CFDEM®coupling 7.1.0 now includes automatic detection of the GPU environment.
Standard run tools, such as cfdemSimulate and Autorun, are now GPU-aware and will automatically launch simulations with the correct configurations.
Additionally, the software now features:
- Significant improvements to environment loading speeds.
- A "zero user input" default setup to get simulations running immediately.
- More concise and helpful error messages to assist in troubleshooting.
Expanded Validation and Thematic Use Cases
To help you build your models with confidence, version 7.1.0 comes with a suite of well-documented cases.
These range from fundamental Validation Cases, such as prismatic fluidized beds, bubble columns, and particle settling to complex thematic Use Cases. New use cases showcase the solver's versatility in applications like spray granulation, erosion around obstacles, and particle separation in flow.
