Low rank approximation (PCA) with MKL's dgesvdx

Question

I have a double[][] A representing a (large) N × N matrix, say N=10,000. I know that this matrix is A=U*transpose(V)+noise, with U,V being N×2 matr