In this study we consider the implementation and application of domain decomposition and generalized gradient iterative algorithms for solving coupled viscous flow and heat transfer problems. The main algorithm development involves parallelizing the computationally intensive kernels for the matrix-vector product, transpose matrix-vector product and dot products for distributed massively parallel systems such as the T3E. Special care is taken to maximize communication efficiency by overlapping interface communication with subdomain interior computations. Performance results and scalability studies on the Cray T3D and T3E will be presented together with flow and heat transfer simulations for natural convection and Rayleigh-Benard examples.

* Speaker