Topological defects form at cosmological phase transitions by the Kibble mechanism, with cosmic strings—one-dimensional defects—being the most studied example. A rigorous analysis of their astrophysical consequences is limited by the availability of accurate numerical simulations, and therefore by hardware resources and computation time. Improving the speed and efficiency of existing codes is therefore important. All current cosmic string simulations were performed on Central Processing Units. In previous work we presented a General Purpose Graphics Processing Unit implementation of the evolution of cosmological domain wall networks. Here we discuss an analogous implementation for local Abelian-Higgs string networks. We discuss the implementation algorithm (including the discretisation used and how to calculate network averaged quantities) and then showcase its performance and current bottlenecks. We validate the code by directly comparing our results for the canonical scaling properties of the networks in the radiation and matter eras with those in the literature, finding very good agreement. We finally highlight possible directions for improving the scalability of the code.