I was not aware that NVIDIA published early this year the white paper
High Quality DXT Compression using CUDA, in which in REF 6 of the paper they kindly reference one of my tutorials on eigenvalues and eigenvectors:
Written by Ignacio Castano, the NVIDIA paper describes an image compression technique using, eigenpairs theory, the power method, and other standard numerical techniques. The way they’ve worked around some “technicals” is quite interesting. A great example of linear algebra in action.