Parallelization of a relaxation method to run on the Intel Paragon

We present a fast parallelized under-relaxation iterative algorithm applicable in solving a large scale Laplace equation that is incorporated into injection molding analysis. The rate of speed gain versus number of processors is studied using three models with different data sizes. Experimental results of running on 32 Intel Paragon processors show reductions up to 98% of original computation time obtained by using a SGI Challenge. Moreover, due to a better memory utilization, the observed speed gain is greater than the proportional increase in the number of processors used.