Advecting 10.000.000 particles

   16870   10   5
User Avatar
Member
1110 posts
Joined: 8月 2008
Offline
I just had to test out that advection thing with tons of particles, and here is what I got so far.

http://vimeo.com/17095369 [vimeo.com]

Big thanks to Peter Quint and his tutorials on Vimeo !

/M
/M

Personal Houdini test videos, http://vimeo.com/magnusl3d/ [vimeo.com]
User Avatar
Member
160 posts
Joined: 1月 2006
Offline
great look)
User Avatar
Member
2 posts
Joined: 4月 2011
Offline
looks great
hardware configuration, simulation/rendering time, if I could ask?
User Avatar
スタッフ
2540 posts
Joined: 7月 2005
Offline
I saw your comment in the vimeo list about accurate births being the longest compute component in your particle simulation.

Have you tried the now-common technique of taking your input geometry and using the Time Shift SOP in combination with a For Loop SOP to create multiple copies of the geometry in-between frames and birth from that?

I have my own asset that I use all the time now. Dead simple. The Time Shift SOP even supports re-timing of volume primitives, so anything input goes.

See the attached example file to pick up the asset. I rarely use accurate births any more.

10M particles? I did a quick 10 wedge pass at 3M advected particles per pass across 240 frames (30M particles) in 2 hours then rendered with each pass as a geometry delay load procedural in Mantra render as points in another 3 hours and with 12Gb of memory. Left a LOT on the table to go much higher.

I'll be showing this workflow setup from scratch (of course) along with a few others at the FMX masterclass on simulations this week. I try not to pluck through pre-built scene files when it's just as fast to construct the network and show the steps.

Attachments:
emitter_interframe_copies_v001.hip (84.9 KB)

There's at least one school like the old school!
User Avatar
Member
2624 posts
Joined: 8月 2006
Offline
Jeff ,
is the master class going to be recorded ?

Rob
Gone fishing
User Avatar
Member
577 posts
Joined: 11月 2005
Offline
another way, to avoid accurate births, would be that one, first and second post

http://forums.odforce.net/index.php?/topic/12751-martins-sketchbook/ [forums.odforce.net]
User Avatar
Member
1110 posts
Joined: 8月 2008
Offline
Cool ! defently something I will be using instead of accurate birth.

BTW this is a 5 old month post, since then I did a 25mil test aswell here:

http://vimeo.com/17234865 [vimeo.com]
/M

Personal Houdini test videos, http://vimeo.com/magnusl3d/ [vimeo.com]
User Avatar
Member
1110 posts
Joined: 8月 2008
Offline
I compared these solutions, here is a file showing the result. Maybe I did something wrong but the last solution posted held up better in my test.

Attachments:
SubFrameFake_Compare.hipnc (142.1 KB)

/M

Personal Houdini test videos, http://vimeo.com/magnusl3d/ [vimeo.com]
User Avatar
Member
8766 posts
Joined: 7月 2007
Offline
it's because you are using $F in your expression for sphere movement (in all sturb() expressions)
and therefore the interframe interpolation is stepped resulting in incontinuous particle emissions

just make sure to use $FF in your expressions
Tomas Slancik
FX Supervisor
Method Studios, NY
User Avatar
Member
1110 posts
Joined: 8月 2008
Offline
Aha yes, worked much better with $FF, and around subframes 35+ i starts to be hard to tell the 2 solutions apart.
/M

Personal Houdini test videos, http://vimeo.com/magnusl3d/ [vimeo.com]
User Avatar
Member
121 posts
Joined: 7月 2008
Offline
Hey guy's great stuff, thanks for sharing.
Available for new projects
  • Quick Links