We did this for a work project recently using a major automation supplier's software. Even with the most powerful Xeon+RadeonRTX machine we could buy, and a very simplified model, it took 100x real time to run. So I get why NV wants to go this direction.... it's not useful without extreme computing power.
The software is necessarily insanely complicated too and it took a team of high end of people to build (or worse, manage) the models. It was cool what it could do if you could figure out how to make it work though.
In the end we just scrapped it and asked for our money back. For what we were doing it would have been cheaper to rebuild the machines from scratch than screw around with the modeling.
Eureka! -- Archimedes