And you can use a spoon to chop spaghetti into pieces. Say I have a 10,000 by 100 matrix (1M elements), squaring that to get a 10,000 x 10,000 (100M elements)

And you can use a spoon to chop spaghetti into pieces. Say I have a 10,000 by 100 matrix (1M elements), squaring that to get a 10,000 x 10,000 (100M elements) matrix to get the singular vectors is probably a bad idea. That squares the condition number too. The SVD staying in Reals when you have Real data is a nice feature.