What is difference between standardization (z score) and normalization (Max-Min technique) of features for PCA?

What is the correct (or better) way to preprocess the data for training?

1) Normalize the data -> PCA -> training

2) PCA -> normalize PCA output -> training