Looking for a source to cite for the fact that conjugate gradient converges more slowly when matrix is positive semidefinite

Question

This is surprisingly difficult for me to look up for some reason. I assume it's stated somewhere in a canonical textbook on the subject but I haven't been able to find anything. For a simple example, let's say the energy function we're trying to minimize is $(x-y)^2$. Then the matrix for this function is $A = \begin{bmatrix} 1 & -1 \\ -1 & 1 \end{bmatrix}$, which is positive semi-definite (has an eigenvalue of zero corresponding to a translational symmetry). I'm just looking for a source for the fact that this system will converge more slowly than a similar system whose matrix is positive definite will, due to this translational symmetry.