Cardiac magnetic resonance imaging (cMRI) is an integral part of diagnosis in\nmany heart related diseases. Recently, deep neural networks have demonstrated\nsuccessful automatic segmentation, thus alleviating the burden of\ntime-consuming manual contouring of cardiac structures. Moreover, frameworks\nsuch as nnU-Net provide entirely automatic model configuration to unseen\ndatasets enabling out-of-the-box application even by non-experts. However,\ncurrent studies commonly neglect the clinically realistic scenario, in which a\ntrained network is applied to data from a different domain such as deviating\nscanners or imaging protocols. This potentially leads to unexpected performance\ndrops of deep learning models in real life applications. In this work, we\nsystematically study challenges and opportunities of domain transfer across\nimages from multiple clinical centres and scanner vendors. In order to maintain\nout-of-the-box usability, we build upon a fixed U-Net architecture configured\nby the nnU-net framework to investigate various data augmentation techniques\nand batch normalization layers as an easy-to-customize pipeline component and\nprovide general guidelines on how to improve domain generalizability abilities\nin existing deep learning methods. Our proposed method ranked first at the\nMulti-Centre, Multi-Vendor & Multi-Disease Cardiac Image Segmentation Challenge\n(M&Ms).\n