École Nationale Supérieure de
Techniques Avancées

Résumé

This paper introduces a compression method using Discrete Wavelet Transform (DWT) to address the challenges of limited GPU memory capacity and slow data transfer rates between CPU and GPU in large-scale fluid simulations. The proposed approach allows for the storage of entire fluid simulations in compressed form on the GPU, minimizing the need for frequent data transfers. By dynamically decompressing only the required portions during simulation, we achieve a balance between compression ratio and simulation accuracy. Extensive experiments demonstrate the method's effectiveness in overcoming the limitations of GPU memory capacity and data transfer rates, offering a promising solution for high-fidelity fluid simulations on resource-constrained GPUs.