Operating system: Antarctic Waterfowl Delight (Linux)
Graphics card: Intel
Short description of error
Instead of imparting the same impulse per volume per time delta, the "Fluid Density" setting for cloth simulations seems to impart the same impulse per volume per simulation step.
The result is that cloths with higher "Quality Steps" will fall more quickly.
The two Suzannes have exactly the same cloth sim settings, excepting that the one on the left has 5 for "Quality Steps" and the one on the right has only 1.
A lot of instability and slight accumulated positional variance is of course to be expected with different step sizes. But my understanding is that the physical forces in the sim should be scaled (and most of them are scaled) so they behave consistently with scene time or scene frames, not with simulation steps.
Setting the density to 0.0 does indeed produce much more time-consistent results, even though the quality steps settings are still different:
Exact steps for others to reproduce the error
- Enable cloth simulation on a solid/manifold mesh object.
- Set and enable a non-zero fluid density for the sim.
- Duplicate the object.
- Give the duplicate's cloth sim a different "Quality Steps" setting, but leave everything else the same.
Or download the attached file: