Raster GISes like GRASS, ArcGIS/Spatial Analyst, and Idrisi can perform a rich set of data processing and analytical procedures loosely known as "map algebra." In today's computing environment it is becoming commonplace to maintain rasters of 100,000,000 cells or more in many different formats and to demand relatively complex calculations like viewsheds, watersheds, and terrain identification, as well as image processing capabilities.
It seems that many open source, free, and inexpensive solutions are out there. But which ones really hold up in practice? That is, which ones can handle large grids efficiently, can easily get data in and out, are reasonably bug free, and offer a full complement of analytical procedures? What are the pitfalls or hidden limitations that you don't find out until you have invested lots of time in learning these systems? (This last question is one that is not easily answered with Web searching and where I am hopeful respondents can offer valuable advice.)
I am especially interested in solutions that can both integrate well with and compete with the popular (but expensive) commercial systems (which means Windows compatibility is important).