I'm looking into if cloud-computing services (e.g., AWS but others as well) can feasibly supplement the traditional clusters (such as Comet, Stampede2, Pleiades) I've been using to solve fluid dynamics problems.
That is, I'd need to run my code (not OpenFOAM which AWS advertises on one of their plans) and it would need ~1TB RAM ~360 processors which have a fast interconnect (and are in the same location to minimze latency).
The more general question is: does cloud-computing mean that everything is virtualized on some unknown, possibly small machine, or can one request use of specific hardware using cloud computing (where in this case, it's being used in the same way as a traditional cluster but the allocation process may be rather different)?