Реалізація матричного множення з використанням топології Mesh Network на основі DirectCompute

Abstract

The paper considers the implementation of a matrix multiplication algorithm using the Mesh Network topology and DirectCompute technology. The features of a hybrid computing architecture are analyzed, where the central processing unit performs control functions, and the graphics processing unit is used for massive parallel computations. The feasibility of using a mesh topology for modeling streaming computations and pipeline execution of compute shaders is substantiated. A software system is developed as a console application in C# using the .NET environment, which simulates DirectCompute mechanisms and CPU–GPU interaction. The proposed solution demonstrates the effectiveness of parallel data processing for matrix multiplication tasks and can be used as a training model for studying modern GPU computing approaches.