The MetService investigated the performance of the numerical weather prediction model Weather Research and Forecasting (WRF) on the Pan cluster in Auckland and the Bluegene/P and AIX in Christchurch, respectively, through a New Zealand eScience Infrastructure proposal development. The weather model was executed in different configurations running different resolutions and parallelised over different amounts of hardware, respectively. The interesting results from these experiments are presented, in particular the performance with regard to cost effectiveness, speed, and scalability. The presentation will also mention the support received from the centres and the upfront cost of getting the experiments going. Another item we will discuss is the suitability of these HPCs with regard to both operational delivery of and research into numerical weather prediction with WRF. Also the performance is compared with the cost and scalability of other options of generating the same research data.