Scalable/Iterative Large Data Frame Dimensionality Reduction R

Question

I often have truly large data frames (ie 10 to 40 columns, millions to hundreds of millions of rows) that I would like to perform dimensionality reduction on in R/CRAN. Projects are Remote Sensing / Raster / Classification / Regression problems. I have around 170GB of ram, but most of the algorithms I've tried still frequently exceed this on most projects. Are there sample based or incremental / batch packages I should be aware of, or other standard approaches I'm missing? I understand that it may be possible to calculate the necessary statistics incremental my self, but I'm surprised I can't find a package ready made to do this, as I dont believe this situation is rare. I am aware of some GIS/Remote Sensing Software that have options (Orfeo, Arcgis, ENVI, ERDAS etc...) for raster based solutions. I'm most interested in techniques outside of PCA.