Performance improvement?

[zforest]

I'm testing with the human model.

I made two of them as ObiSoftbody into a scene and drop them from a height. I get inaccurate collision as they overlap each other.

So I go to either decrease the particle radius to 0.05 from 0.1, to make the particles generated more dense. But then the simulation drops to a crawl.

I also tried changing the overlap to 2x of default.

What am I doing wrong? The Octopus demo video has a dozen such softbody models and perform great.

I have a i7 8700 cpu

Thanks

[josemendez]

I'm testing with the human model.

I made two of them as ObiSoftbody into a scene and drop them from a height. I get inaccurate collision as they overlap each other.

So I go to either decrease the particle radius to 0.05 from 0.1, to make the particles generated more dense. But then the simulation drops to a crawl.

I also tried changing the overlap to 2x of default.

What am I doing wrong? The Octopus demo video has a dozen such softbody models and perform great.

I have a i7 8700 cpu

Thanks

Hi there,
- Try using less particles, and increase particle overlap. This will close gaps in the particle representation without the need to use more particles.
- Keep the cluster radius as small as possible. Large clusters with many particles will cause a performance hit.
- Tru using more particle collision iterations. This will improve convergence speed (quality) for particle-particle collisions alone.
- Try using 2 or 3 substeps. This will greatly improve overall convergence speed and simulation quality, allowing particles to pick up collisions more accurately.
- Try reducing your fixed timestep. This has a similar effect to increasing the amount of substeps.

Iterations and substeps are probably the most important parameters to understand. They largely determine the simulation quality/speed ratio. To see how iterations and substeps have an impact on the simulation, this is a good read: http://obi.virtualmethodstudio.com/tutor...gence.html