The recent performance increases in graphics processing units (GPUs) have made graphics cards an attractive platform for implementing computationally intense applications. With their numerous parallel computational pipelines and SIMD architecture, modern GPUs can outperform high-end microprocessors by one to three orders of magnitude, depending on the problem. Most work to date at EM Photonics and elsewhere has focused on accelerating specific applications by porting core engines onto the GPU. In this project, we propose the development of general purpose computational libraries capable of solving numerous core numerical functions on commodity graphics cards. These solvers will be based on accepted, industry-standard interfaces and will be easy to integrate with current and future applications. The result will be a GPU-based numerical coprocessor capable accelerating a wide range of computationally intense functions, thereby reducing processing times in applications where numerical computations are the primary bottleneck.