This article lists the solution methods in Feko that support GPU acceleration for sequential (single CPU core) runs.
Feko supports the use of a graphics processing unit (GPU) for simulation acceleration using the unified device architecture (CUDA) framework from NVIDIA.
The computational phases targeted for execution on a CUDA-based GPU show a significant speedup compared to standard CPU-based execution.
Note: Minimum requirements for the CUDA device:
- Compute capability of at least 7.0.
- Driver installed on system must support CUDA 12.8.
The following solution methods support GPU acceleration:
- MoM* for single and double precision matrix solve (LU decomposition): Support single or multiple GPUs.
- CMA: Supports multiple GPUs.
- RL-GO for manual (fixed) ray launching grid and flat meshes (not curvilinear): Supports multiple GPUs.
- FDTD: Supports single GPU.
* Includes SEP, VEP, planar Green's function (multilayer substrates) and thin dielectric sheets.
