Методи створення й опрацювання 3D-моделей з використанням CUDA-технологій

Abstract

The paper provides an analysis of methods for accelerating computationally intensive tasks of creating and processing 3D models. The characteristics of the proprietary NVIDIA CUDA platform are investigated, and its comparative analysis with the OpenCL open standard is conducted in terms of performance, programming model features, and efficiency of using the GPU hardware architecture. A comparative analysis methodology is proposed by developing software modules that implement identical parallel algorithms for geometry processing and physics simulations. Recommendations for optimizing data access using the GPU memory hierarchy (shared/local memory) for 3D structures are developed. The research results show that CUDA demonstrates higher performance and has a more mature ecosystem for NVIDIA GPUs, whereas OpenCL provides hardware independence and software code portability.