SLIM ocean model runs 50x faster with multi-GPU, rivals 1500 CPU cores

Unstructured-mesh ocean models like SLIM are critical for coastal applications due to their ability to handle complex geometries and local grid refinement. However, the Discontinuous Galerkin finite element (DG-FE) method used in SLIM is computationally expensive, often limiting its broader adoption. A team of researchers from UCLouvain has now implemented a full 3D DG-FE ocean model optimized for both single- and multi-GPU systems, targeting NVIDIA and AMD architectures. Through memory layout optimization, kernel-level parallelization, and matrix-free solvers, they achieved dramatic performance gains. A single HPC-grade GPU (e.g., NVIDIA A100) delivers performance equivalent to approximately 1500 CPU cores. Replacing a 128-core CPU node with a 4xA100 GPU node yields a speedup of around 50x, and weak-scaling efficiency is maintained up to 1024 GPUs.

To demonstrate real-world utility, the team applied the GPU-accelerated SLIM model to the Great Barrier Reef. They achieved a spatial resolution five times finer than the most accurate existing model while maintaining a physical-to-numerical time ratio of 100—meaning the model simulates 100 days of ocean dynamics per day of computation. This breakthrough makes ultra-high-resolution coastal simulations feasible for the first time, opening doors for more accurate storm surge predictions, pollution tracking, and climate impact studies. The work underscores how GPU-based HPC can overcome the computational barriers of DG-FE methods, particularly for large-scale environmental modeling.