Very High Overfitting Upper Bound Value for Hidden Nodes

Hello,

I am trying to implement regression/curve fitting NN. My dataset is I = 1, N = 52500, O = 3.

I would like to find out NMSE and Rsq values for different number of hidden nodes in a loop. My loop will execute between 0 and No overfitting upper bound value. I am computing the No overfitting upper bound value using this formulae,

H =((N*O)-O)/(I+O+1).

In my case, the value is 31500.

Can I use any other upper bound value without affecting the network performance?

Any help would be great. Thanks.