Cases where NTK and NNGP are very different

[YehonatanAvidan]

Hello everyone,
I'm looking for types of data sets where the NTK prediction and the NNGP prediction will be very different.
Both I'm calculating according to theory using k(x_test,X_training)*inverse(K(X_training, X_training)) *Y.
Where Y is the labels vector, k test is the kernel between the test point and the training and X is the training matrix).
For NTK I use NTK kernels and for NNGP I using NNGP kernels.
Is there a problem set in which these values will be very different