dt (timestep)
Smaller is more accurate but slower. If the simulation blows up, lower this first.
ω (SOR relaxation)
Controls pressure solver convergence speed. Values near 1.9 are optimal; above 1.98 can diverge.
Iterations
Pressure solver iterations per frame. More iterations means better pressure accuracy but slower frames.
Inflow velocity
Speed of the incoming flow — the U in Re = U·D/ν. It does not raise Reynolds number: the Re control sets ν = U·D/Re, so raising the inflow raises the viscosity with it and Re holds constant. It does raise the CFL number, and it takes you off the one amplitude (U = 1.0) at which the solver's numerical viscosity was measured — so the Re badge stops quoting a measured ceiling once you move it.
Grid resolution
Number of cells across the domain. Higher resolution captures finer detail but demands more GPU work. It does not buy Reynolds number. The scheme's own numerical viscosity was measured independent of cell size — flat to 0.5% across a 4× refinement — because it is an operator-splitting error in time, not grid diffusion; only a smaller dt lowers it. And at the 256 pressure iterations the Kármán preset ships, the honest Re ceiling falls as the grid grows — Re 236 at tier 64 down to Re 16.5 at tier 1024 — because a fixed iteration count converges the pressure solve progressively less well on a finer grid.
▸ Learn more about numerical stability
The solver uses a MacCormack advection scheme — a semi-Lagrangian trace forward, the same trace backward, then a limited combine that keeps the correction only where it stays inside the min/max of the cells it was interpolated from. For stability, the CFL condition (dt × U / h < 1) should be satisfied — meaning fluid should not cross more than one cell per timestep. The default parameters are chosen to be stable for typical configurations. If the simulation explodes into noise, lower dt or reduce inflow velocity.