This work has concentrated on the serial optimisation of a number of routines in DL_POLY_3. The routines were identified by profiling and the work has resulted in a roughly 10-25% increase in performance dependent upon the number of cores in use. Given that the main driver for this project was ultimately the poor performance of DL_POLY on the XT6 compared to the XT4, to have the code now running on the XE6 appreciably faster than on the original architecture is a very satisfying result.

Valène Pellissier 2011-08-24