Recent advances in high-throughput genomic technology, such as micro arrays, usually produce vast amounts of gene expression data under many experimental conditions. Analyzing such data is often difficult due to the colossal data size and the intensive computing involved. In addition, many existing analysis tools often require the inference of experienced analysts and subjective judgments. In this paper, we developed a parallel approach based on Random Matrix Theory (RMT) to generate transcription networks using Graphical Processing Units (GPUs). Recently, GPUs have been redesigned into a more unified architecture, which has allowed them to be used more readily in general purpose computing. This architectural advancement has resulted in GPUs becoming easily programmable parallel processors with performance that is vastly superior to CPUs. Our GPU-based approach makes automated micro array data analysis faster, more accurate and noise resistant without engaging remote high performance computing facilities, such as a cluster or supercomputer. The implementation moves some computationally intensive tasks, such as the calculations of Pearson correlation coefficients, tridiagonal reduction, back transformation of eigenvectors, and orthogonal rotation, to the GPU. Experimental results on real micro array datasets show that our GPU implementation runs faster than a CPU version using highly optimized LAPACK routines. The runtime speedup gets higher as the number of genes and sample points in a micro array dataset increases.