A MODERN VIEW ON GPU ARCHITECTURES AND DATA-PARALLEL COMPUTING FOR DATABASE AND SIGNAL PROCESSING APPLICATIONS

Abstract

Abstract: The rapid growth of data-intensive applications in computer vision, database systems, and digital signal processing has significantly increased the demand for high-performance computing platforms. Graphics Processing Units (GPUs), originally designed for graphical rendering, have evolved into powerful general-purpose parallel computing devices capable of accelerating a wide range of computational tasks. This paper presents a comprehensive analysis of modern GPU architectures and their applicability to data-parallel problems in database management and signal processing. Special attention is given to architectural evolution, parallel programming models, memory hierarchy, and algorithmic adaptation strategies. The study demonstrates that the effective utilization of GPUs enables substantial performance improvements for large-scale data processing tasks, particularly in applications characterized by high degrees of parallelism. Experimental observations confirm the efficiency of GPU-based solutions in reducing computational complexity and improving throughput compared to traditional CPU-based approaches.