clustering of a 2D interaction matrix

I have a 2D square matrix of interaction (M[y]= an integer value representing the number of interactions between x and y (x!=y). x and y are at the same time coordinates in the matrix and label that define a certain point). The matrix can be quite big (e.g. 60000*60000)
I’d like to cluster it (regroup the points that interact with each other or through another point, and of course keep track of the initial coordinates to know who interact with who)
Are there CUDA libraries to do that?
If not does anyone knows an algorithm that i could implement?
Thanks a lot

Look for some GPU/CUDA applications in BioInformatics.
There should be some tools for this already …