3D PET filtered backprojection image reconstruction is computationally intensive requiring twenty-five times the computational time over 2D PET using the same computer hardware. In addition, the EM algorithm requires between two and three orders of magnitude times the computational time of the filtered backprojection algorithm. Our approach to meet these computational demands is the development of a network based computational workstation providing service for all PET systems at UCLA. The computational server is constructed from the integration of board level products from multiple vendors. It is our design goal to provide two orders of magnitude increase in computational power over our existing computer hardware, to service the 3D reconstruction needs of the differing PET scanners, and reconstruct the various 3D PET protocols. In addition, the EM algorithm is investigated first with 2D PET data on this architecture and then the 3D PET case is discussed. Here we report on the workstation hardware and software design, and on its construction.