Abstract
In the talk, we briefly explain modern PC architectures and the general principles of heterogeneous computing. We illustrate how multi-core and GPU computing may be utilized for higher performance and more robust VRP solvers, and explain the details of our solution method for the DVRP. We present the results of computational experiments on standard CVRP/DVRP benchmarks from the literature as well as industrial test instances from newspaper distribution. Perspectives and directions for future work are given.