You can try it out, whats in memory is a compressed mathematical map of your data, typically it uses about 30-34x less memory than the total dataset size.
Currently the only ram hungry part is during the graph build (which we're working on lowering too.)