Are there any useful tips that can help me reduce/optimize the memory footprint required to store a huge Eigen sparse matrix? For example, are there useful icc compiler flags, or would specifying apriori the number of nonzero elements yield a better performance?

We have two sparse product implementation, the default one:

R = A * B;

and a second one:

A = (A*B).pruned();

which can in addition remove small entries on the fly:

A = (A*B).pruned(eps);
A = (A*B).pruned(eps,ref);

see this page:
http://eigen.tuxfamily.org/dox-devel/Tu ... FeatureSet