Performance of PGA (Programmable Graph Architecture) for Matrix Multiplications

Matrix multiplication is one of the primary operations in linear algebra and is useful in a wide spectrum of multimedia applications including signal and image processing. In this paper we review a novel computer architecture for matrix multiplications. This novel architecture is based on graph theory, hence the name programmable graph architecture. We discuss the performance of PGA via time and space complexity and time-processor product for parallel implementation. We further compare the PGA performance with Cannon´s algorithm and show that the PGA architecture outperforms the existing method for matrix with any dimension