Steven,
The locations we want accurate stresses calculated are a small proportion of the total model.
We normally have to run with a 'raw' mesh first to get a 'feel' for the behaviour
Then we home in with mesh refinements.
The rest of the model is only there to provide the correct load paths into and stiffness surrounding the location(s) of interest.
Sometimes the part/assy is too big for this 'global' model approach with mesh refinements and an increase in the amount of RAM would be welcome but ...
I have this feeling that presenting users with the ability to use much more memory then they would simply fill it with unnecessary elements rather than being efficient; the quality of FE models would not really be enhanced.
Wouldn't sub modelling be a more strategic development first?