An HPC friend altered me to a recent announcement of a new memory-optimization product. The article in HPCwire is here: <http://www.hpcwire.com/features/Swedish_Startup_Looks_to_Unleash_Multicore.h...> and the company website is here: <http://www.acumem.com/>. Note sure of something like this would be useful in the numpy conext, but I thought I would point it out in case it was and the right people in the numpy community weren't aware of it yet.