raymondu999 wrote:Forgive my ignorance, but I used to do something similar-ish for submarine propulsion; wouldn't these calculations, though you have millions/billions of them, wouldn't each individual calculation be actually rather simple? I don't remember the kind of computational power I was using, but I think for a golf ball (never did something that small before, so I could be wrong) that 670 procs would be overkill, no?
If you are doing DNS at high Re computational power is never enough because you are not modelling turbulence, but directly calculating it (Direct Numerical Simulation). So you need much finer spatial grids and shorter time intervals.
It is the top level of simulation hierarchy (which tells: RANS, URANS, DES, LES, DNS), the most demanding in terms of computational effort.
F1 teams instead often use thousands of processors for performing RANS / URANS job quickly with meshes around 100-300M cells.
Maybe some of them have already approached DES and LES