This paper presents the design and evaluation of a practical nonlinear precoder for downlink MU-MIMO operation in a 5G system with a large array. The proposed precoder combines conventional linear precoders with a Tomlinson-Harashima precoder (THP) to achieve a lower complexity implementation, and provide more flexibility for various types of CSI feedback while offering substantial performance gain. Through extensive and realistic simulations, we provide performance evaluation of the proposed design and benchmark its performance against widely adopted linear precoders. Our numerical results demonstrate that the proposed precoder can achieve gains in sum rate over linear precoders even with limited and delayed CSI. Furthermore, the performance of the proposed scheme is more robust to smaller array sizes and various propagation conditions.