LSTM behaviour of h_0 and c_0

When using a batch is it using a memory of h_i and c_i over all the batches?
So when evaluating, do you get different results when using batches vs not?