GPU accelerated phase-field simulation of convection effect on dendritic growth

A phase-field model incorporating mass and momentum conservation equations for binary alloy is established. Due to the problems of large computational demand and low computational efficiency, it is hard to implement real-time phase-field simulation with traditional methods. In this paper, based on CUDA techniques, parallel solution to dendritic growth of Ni-Cu binary alloy under convection is implemented. A speedup of 46.52 is achieved on a single GeForce GTX780 GPU computation with the shared memory compared to a single Intel Xeon E5507 CPU computation when the grid size is 5123, which greatly improves computational efficiency. In addition, the effects of convection on dendritic pattern and microsegregation are also investigated. The results show that, without noise, the convection can induce side-branches and the upstream side-branches are more developed with larger convection velocity U. And an increased level of convection velocity U aggravates the severity of dendritic microsegregation in the front interface. Copyright © 2015 American Scientific Publishers.