Exponential slowdown on large particle emission

[Gustavo Eggert Boehs]

I was just simulating 3 milion particles and noticed SI doing somethig very odd. If I create millions of particles by generating a sample set and adding points I get them almost instanenously (less than a seccond). Now when I use the regular emit node emmiting 3mil particles can take hours... HOURS.
Other numbers:
100.000 particles = 3 seconds
200.000 particles = 15 seconds
400.000 particles = 121 seconds!!

Anybody else getting this?

[Sebastian Kowalski]

by any chance via volume emission?

[Chris Chia]

How did you setup your icetree?
I have created 3 million particles under 20secs (setting 1 million for Rate in No. of Particles Per Sec)

Chris

[Gustavo Eggert Boehs]

Sorry for missing the specifics guys. (Yes) It was a volume emission, and this behaviour is better perceived when particles are created all at once. It is also important to notice, that once you have created the gazzillion particles, and waited for hours, if you try to do it again it does it all blazingly fast. At any rate, here are the steps to reproduce this:

1. Create a cube

2. With the cube selected, under ICE - Create, choose Basic Emission

3. Go to frame 2

4. Change the emission type to volume

5. Change Rate type to Total Number of Particles

6. Change the rate of particles to a very high number (say 3 million, or increase little by little get a feeling of the exponetional slowdown)

7. Repeat the volume point creation with a basic add point/generate sample set/iniate data and see how fast that works

[Ciaran Moloney]

Hi,

yes volume emission is slow. I guess that's just the nature of the problem, especially since ICE volume emission is really accurate. You can get quicker volume emission by generating lower resolution distance fields that approximate the shape.

I'm just making an uninformed guess here, but I believe Softimage is caching the emission volume, which is why you get an instantaneous update using the generate sample set after having already emitted using the other node. If the volume is already constructed then sample creation is easy. Try emitting from a deforming geometry with time varying emission and you shouldn't notice much difference at all.

Ciaran

--
- Ciaran

[Rob Chapman]

yes, Volume emission has always been slow and never been improved since ICE came out. see Ciairans sdf emit for the speed comparison - think it may be quicker than default add point :)

http://blog.blackredking.org/?p=47 

[Gustavo Eggert Boehs]

The instantaneous update using generate sample set may be observed on a new scene with no previous caching.

Im aware of Ciaran's SDF compound, which is really smart by the way... but my guess is that the emit node is doing something really silly, and I want to point that out, as it is the standard point creation method available in vanilla Softimage.

Fact is the emit node is scaling exponantially while adding points from a sample set scales linearly, I do suggest you guys try this by yourselfs :)

[Ciaran Moloney]

Try plugging the jitter emission along velocity node (taken from emit from geometry) into your add point's execute on emit. Looks like that causes the geometry to be evaluated multiple times (per-particle?)

--
- Ciaran