Rbd And Particles Fluid Collision, Solver Substeps, Fluid Sim Speed ?

[Vinz9]

Hi,

Is there some rule on how to choose substeps when using a rbd solver and a particle fluid solver ?

I did a test with a particle fluid colliding with a wall made of thin boxes, and I had to raise min and max to 10 in the rbd solver if I wanted the fluid not to explode.

I would have thought 1/ only the particle solver substeps would matter and 2/ why do I need to raise min substeps in the rbd solver and not only max substeps ? Isn't it supposed to increase substepping automatically when needed ?

Also how does the time step on the dop network relates to the solvers substeps settings ?

Last question, how can I improve the speed of a particle fluids simulation?

I did a quick comparison with realflow, and for a sphere of particle fluids colliding with a plane with roughly the same number of particles in both apps (about 33000), it took 4-5 times longer in houdini ( with a quadcore q6600, for 50 frames, about 5 minutes in realflow, and 20 minutes in houdini).

Thanks a lot for any suggestion

liq.avi

Edited February 19, 2008 by Vinz9

[eetu]

I did a quick comparison with realflow, and for a sphere of particle fluids colliding with a plane with roughly the same number of particles in both apps (about 33000), it took 4-5 times longer in houdini ( with a quadcore q6600, for 50 frames, about 5 minutes in realflow, and 20 minutes in houdini).

Can't help you with the optimization, sorry.. But RF is multithreaded while dops are (mostly?) not, so with a quadcore you could expect a close to 4 times win in any case :\

eetu.

[Vinz9]

Oh, I thought DOPs were multithreaded. Thanks for clearing things up.

[whalerider]

Don't forget that RF is specializing in fluids and was created several years ago.

Houdini's fluids are a new feature, which will improve over time.

[digitallysane]

DOPs are multithreaded.

[eetu]

Whoops *insert foot in mouth*

As a quick test of a couple of the example scenes: particle fluids took 70-80% and volume fluids 55-60% of processor time on a dual core machine.

Soo, halfway there

eetu.

[Asmat]

Hi,

Is there some rule on how to choose substeps when using a rbd solver and a particle fluid solver ?

I did a test with a particle fluid colliding with a wall made of thin boxes, and I had to raise min and max to 10 in the rbd solver if I wanted the fluid not to explode.

I would have thought 1/ only the particle solver substeps would matter and 2/ why do I need to raise min substeps in the rbd solver and not only max substeps ? Isn't it supposed to increase substepping automatically when needed ?

Also how does the time step on the dop network relates to the solvers substeps settings ?

Last question, how can I improve the speed of a particle fluids simulation?

I did a quick comparison with realflow, and for a sphere of particle fluids colliding with a plane with roughly the same number of particles in both apps (about 33000), it took 4-5 times longer in houdini ( with a quadcore q6600, for 50 frames, about 5 minutes in realflow, and 20 minutes in houdini).

Thanks a lot for any suggestion

No rules, I increase mine enough for the particles to not explode, with my current simulation i had to turn it up to a rificulous 40min, 50 max.

Generally, the substeps are individually controlled by each solver, fluid solver substeps for fluid calculations, rbd substeps for rbd calculations.

I heard particles were not multithreaded, or particle fluids, correct me if im wrong, but arnt level-set fluids multithreaded? (Im on 8 xeon cores, and its hella slow on my end)