Yes atlas is good. For small, dense n you're not going to beat the cubic algorithm. This is the most common case in a lot of applications. For sparse n, use a

Yes atlas is good. For small, dense n you're not going to beat the cubic algorithm. This is the most common case in a lot of applications. For sparse n, use a sparse version of the cubic algorithm. For large dense n, those sub-cubic algorithms might be better. Here n would have to be so large that you'd have to factor in the cost of doing it out of core.