We have rewritten the UCLA AGCM in a form suitable for distributed memory massively parallel processing computers. We have chosen to use Richardson's 1922 two-dimensional domain decomposition as the strategy to achieve parallelism. This technique is well suited to the finite difference hydrodynamics scheme of Arakawa and Lamb as well as for the column based physical parameterizations. The performance of a wide variety of computers, including the IBM SP2, Cray T3D and Meiko CS2 will be compared to that obtained on a single processor CRAY C90. Validation of the rewritten code is complicated by propagation of roundoff differences, principally by the Arakawa-Schubert cumulus convection scheme. Therefore, to validate the parallel code and machines for long term simulations, we are using the climate statistics obtained from 10 year AMIP simulations as the principal diagnostic regarding the correctness of the new implementation. Preliminary results from several runs will be presented.