Can I use a HDD Paging file as RAM for a Giant Fluid Simulation? I know the HDD is much slower than RAM. Maybe if this where true, their would already be examples of it in use.
Would this work for Elbeem or Mantaflow or The Fluid Particle system. Could I still render it with a gpu or cpu when I am done?
This is the article that made me curious. The article is not related to Blender.
I really don’t know, but my intuition is that you’d be lucky if it only gets 30x slower and your normally 24 hour simulation takes a month to run.
But try it and let us know! 
Besides buying more RAM, you could always consider renting a cloud server from Amazon AWS EC2 (or Microsoft or Google’s equivalent) to run your simulation.
Or just turn down the knobs until it fits in whatever memory you have. I can remember a time when 1MB of memory cost $12,000 so your 40GB would have been almost a half a billion dollars in those days, and we figured out how to make do with what we had! Of course I don’t remember running a lot of fluid simulations in the 80’s either.
If you have large RAM, i think you can create a virtual HDD in RAM, then set blender temp dirctionary path to virtual HDD.
RAM read/write speed can reach 17GB/s and a lot more if you have DDR4 memory. Hard drives struggle to get more than 50MB/s in the best case scenario (sequential access), which is 350x slower. An decent SSD’s can reach 200MB/s, but it’s still way too slow to be used as RAM.
So, if you want to use your HDD as RAM, then it’s possible but it would be painfully slow.
But if you meant caching the results to the hard drive (ie: when one frame is simulated, the result will be stored in the hard drive) then it’s how you should do it. As long as you have enough RAM to simulate one frame you should be fine