Jump to content

Why cloth sime so slow with finite element solver?


Recommended Posts

I just make a very sample cloth scene, a grid with remesh to triangle topoloy as cloth object, an animated sphere as collided object, let the cloth fall down by gravity at the beginning then hit the sphere and move with shpere, I have 32 threas of CPU, not very high sub div in remesh node, but the speed is very slow, around 6 second per frame, which in 3dsmax almost have realtime playback. Why?

Link to comment
Share on other sites

have you try to lower the solver parameters

- max substep > 128

- max iterations > 128

- max collision > 4

- max local collision > 0

- disable all fracturing parameters

 

sometime you can also disable self collide in cloth object , it's very fast ...

Link to comment
Share on other sites

I don't think houdini cloth is fast, I think is very slow and nearly unusable.

 

Emmanuel, have you tried any other cloth solver? I'm just curious and trying to understand why I think the exact opposite :mellow:

 

Using low settings and disabling self collision and getting a crappy sim is not always an option.

Link to comment
Share on other sites

well i agree that FEM cloth is not on par with nCloth for speed. But it's usable in production (actually i use it to make liquids :) )

 

i agree (and i ask in the 2015 wishlist) that having a cloth that would only use pop solver exactly like the granular stuff would be a nice addition.

 

self collision in POP is one of the big thing i miss from nParticles. and from pop self collision you just need to add spring connection to get some faster cloth.

 

but the tools houdini offer to play with cloth attribute open a lot of creative possibilities ...

i think i'm gonna try to stay as much as i can in h now. but i am a FX guy not a CFX guy ... and in that regard

- ncloth

- nhair

- + the new sculpting tools in maya

- + maya blendshape

give the edge to maya for cloth on a large scale imo ...

Link to comment
Share on other sites

H14 cloth is way better than H13, for speed and fracturing, aka tearing, but H12.5 still has double the solving performance.

Just in case you haven't and ever find some time to do so, it might be worth while submitting your H12.5 and H14 hip cloth files for comparison. I've heard of cases where people did this but the files were using order of magnitude differences in the solver parameters due to changed semantics.

Link to comment
Share on other sites

Just in case you haven't and ever find some time to do so, it might be worth while submitting your H12.5 and H14 hip cloth files for comparison. I've heard of cases where people did this but the files were using order of magnitude differences in the solver parameters due to changed semantics.

 

 

Good idea - this will force me to quantify it properly!

Link to comment
Share on other sites

have you try to lower the solver parameters

- max substep > 128

- max iterations > 128

- max collision > 4

- max local collision > 0

- disable all fracturing parameters

 

sometime you can also disable self collide in cloth object , it's very fast ...

I just can not found these branche of parameters in new solver...

Link to comment
Share on other sites

  • 2 weeks later...

I'm resurrecting this old topic because there seems to be so little information on the new FEM cloth and the default settings are pretty much rubbish.

 

This has been a helpful topic in setting up my first bit of cloth with the system.  At DD, SPI and R+H it wasn't something that fell into the scope of the VFX.  I'm at a small boutique for the now and I'm it.  Anyway, I've got a character that moves rather briskly out of the frame wearing a hooded cloak and it just wants to break the solver at the first big velocity frame, bad enough to crash Houdini altogether.

 

I was able to work it down to being the contraints that were likely the cause of the crashing.  If I disabled the Cloth Attach Constraints it was able to push on with the collision object (volume collision) ripping through the cloth (no fracturing).  The problem is the documentation is so sparse (most parameters have no more information than what appears in the mouse-over), incomplete (parameters in the interface not mentioned in the docs at all) or old (parameters from an older version appearing in the docs but no longer represented by the interface).  

 

I was eventually able to get "something" out for a complete range by doing three sims with subsequent sims started from the time before the last one broke with a new set of constraints.  But I envision all manner of characters with capes or flowing cloaks and dresses where this simply wouldn't be viable and the whole process can't realistically be this horrible, can it?

 

The settings I'm at are also really, really slow.  Some frames, before crash, topping 2hrs for a bad solve on an i7-5820K @ 3.3GHz while the solve barely ever pushes very far over 20% CPU and I've over 20Gb of memory still available to do other things while the FEM eventually has an aneurysm and dies.

 

4x oversampling on the DOP network, because 2x not only didn't stop a crash but what did solve wasn't as good.

 

20 Min Substep Rate ...lower killed the sim sooner, gave worse results

 

0.0612 Substep Tolerance  ...interaction with substep rate settings and oversampling on DOP Network unclear, just says lower is better

 

4 Max Global Collision

16 Max Local Collision ...seems like I'm more interested in local collision, I don't think I ever tried running it at 1x on the Global setting

 

No self-collision

 

 

...and that's pretty much it.  Lowering two parameters that seemed to most effect the speed were the interaction between the DOP Net oversampling plus the Min Substep Rate.  Going higher might have gotten me a few more frames but I have zero confidence the entire thing would have completed.  Meanwhile there are several other shots that were done by a different artist with the same exact geometry in nCloth and while not great, it always managed to keep the spring constraints attached to the animated figure driving the animation, with some very wild, flying animation and the cloak flowing believably, and could be cached out over a similar range of frames in a manageable time frame.

Edited by pockets
Link to comment
Share on other sites

  • 1 month later...

Turns out the main source of my crashing is likely specifying Volume Collisions.  Switching to the mesh surface I was able to get a sim in another scene with far greater velocity to run without crashing.  Success!  

 

The values I'm having to use to get loose, drooping cloth that doesn't look like it's loaded up with spray starch or made of flexible cardboard are much different than any of the examples offered in the docs, and a couple orders of magnitude different than the mouse-over suggestion that defaults are somehow correct for cloth measured in meters.  But I actually look forward to doing more cloth now rather than cringe at the thought of it.  

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...