I'm interested in putting together a system that combines a moderate number of processors (eg, 2) and cores (eg, 8-12) with a lot of RAM (eg, 100s of GB) to experiment with Neo4j, other graph databases, etc wiki.cfcl.com/Projects/YAGO/Hardware discusses my motivations and one possible approach. If anyone has suggestions on other approaches, I would love to hear about them, discuss their costs and capabilities, etc.