Difference with/without Tensorflow tf.math_ops

Question

Suppose in custom loss function, I want to sum a Tensor x in tensorflow, I can use

tf.sum(x)

but if I use a for loop the result