This chapter demonstrates GPU acceleration of the deep water wave problem. The GPU version solves the identical physics as the CPU version, but uses NVIDIA CUDA for massive computational speedup.
Why GPU Acceleration?
Deep water simulations are computationally intensive: - Multiple FFTs per time step - Each FFT costs \(O(N \log N)\) operations - Thousands of time steps needed - Nonlinear products require transforming to/from physical space repeatedly
GPU advantages: - Specialized hardware for FFTs (CUFFT is highly optimized) - Massive parallelism (thousands of cores) - High memory bandwidth - Typical speedup: 10-100× compared to CPU for this problem
We reuses data structures from the CPU version of Section 4 but reimplements the computation loops for GPU execution.