SummaryThe new International Telecommunication Union Telecommunication G.fast standard for next‐generation ultra‐broadband digital subscriber line systems has stipulated mandatory use of vectoring for crosstalk mitigation. In this paper, we study the performance of downstream precoders in G.fast networks utilising the extended bandwidth up to 212 MHz. In particular, the issue of strong crosstalk encountered in practical copper cable bundles at high frequencies giving rise to channel matrices which are not row‐wise diagonally dominant (RWDD) as predicted theoretically, as well as the ramification of deploying precoders designed based on the RWDD assumption in practical G.fast channels are studied. To serve these purposes, an enhanced stochastic far‐end crosstalk (FEXT) model which provides distinct FEXT transfer functions with the dispersion range characterised experimentally whilst encompassing the dual‐slope FEXT encountered in non‐RWDD cables is used for evaluating the performance of precoders in diverse G.fast deployment scenarios. Furthermore, the performance sensitivity of precoders under realistic amount of channel estimation errors is investigated on a common framework, allowing a direct and fair comparison between linear and non‐linear precoders. Results show that the nonlinear optimal precoder outperforms linear precoders in all channel conditions; however, the former is found to be slightly more sensitive to channel estimation errors on shorter loops. Copyright © 2016 John Wiley & Sons, Ltd.