An Evaluation of Cost Effective Parallel Computers for CFD

An evaluation of the current status of two novel clusters, running Linux and NT, has been made to assess their performance for parallel CFD calculations. The core algorithm of the code is the conjugate gradient method used to solve the pressure equation. For good performance, a low latency and high bandwidth is required and the results indicate that, without substantial investment in this area, neither system is competitive on parallel performance when compared to more traditional distributed memory machines. At present, they are very suitable for development platforms and, for particular applications that are not too sensitive to latency and bandwidth, they offer a very cost effective computing resource.