Implement the surface flow component in Taichi to compare with the existing Cython implementation.
Potential upside:
Potential downside of GPU implementation is additional memory transfer from/to the accelerator from/to the RAM, especially in case of variable input data (typically rainfall), saves to disk, and to other non-GPU components (hydrology, swmm, etc.).
Unified memory architectures like Apple Silicon might be at advantage here.
Implement the surface flow component in Taichi to compare with the existing Cython implementation.
Potential upside:
Potential downside of GPU implementation is additional memory transfer from/to the accelerator from/to the RAM, especially in case of variable input data (typically rainfall), saves to disk, and to other non-GPU components (hydrology, swmm, etc.).
Unified memory architectures like Apple Silicon might be at advantage here.