The software renders of yore actually got me thinking about them in a modern context, how they could be extended, and in what ways they could compete with GPU-based algorithms on a per-cycle basis. I think there is still much area to be explored here, given we mostly stopped researching realtime software rendering as GPUs became dominant - I think there is a lot of unexplored area left.

Also one piece of software I think is severely underrated is Ken Silverman's Voxlap: http://advsys.net/ken/voxlap.htm

Released in 2002, it was way ahead of its time.