large scale smoke

[Ivaylo Katev]

one coworker made it

[ivaylo.ivanov]

Real large scale smoke

[Svetlin.Nikolov]

It's got varying colors and Wilson clouds... and lighting from the sky. We could do some of these..

[TSR_trix]

Svetlin cant hold down horses

[Svetlin.Nikolov]

Haha, nope

[lander]

Actually, I don't think he can make a vulcano eruption looking that good

[Svetlin.Nikolov]

Wanna bet? :P

[Morbid Angel]

one coworker made it

So I am curious, if I had to do a sim like that, but that runs for 1500 frames, picture a rocket take off / trail, what would the sim requirements would be? I see these awesome sims that only run 25 frames and stop. Just curious.

[ivaylo.ivanov]

The trick will be to control the grid size, since it's easy to run out of memory. Otherwise it's just a matter of time and disk space. Rocket is difficult since it creates a huge volume which forces you to use lower resolution. We have some basic support for camera fitting, and maybe the only way is using it to clamp the simulation inside smaller visual region. Far camera can be achieved with a separate low res simulation.

[Ivaylo Katev]

So I am curious, if I had to do a sim like that, but that runs for 1500 frames, picture a rocket take off / trail, what would the sim requirements would be? I see these awesome sims that only run 25 frames and stop. Just curious.

i would recommend separate simulation for launching rocket, in single simulation the result would be worse.
the problem is the contradiction between the jet and the smoke quality. the jet requires high spf value , see the animation explaining the spf meaning , it shows exactly a simulation similar to rocket engine. you can see at spf 8 the jet is good, but the smoke is too smooth.
i would also recommend to simulate two or tree separate side smoke clouds, if you try to catch all the clouds in one sim, this will be too big in size and too many empty area will be covered.

[Morbid Angel]

i would recommend separate simulation for launching rocket, in single simulation the result would be worse.
the problem is the contradiction between the jet and the smoke quality. the jet requires high spf value , see the animation explaining the spf meaning , it shows exactly a simulation similar to rocket engine. you can see at spf 8 the jet is good, but the smoke is too smooth.
i would also recommend to simulate two or tree separate side smoke clouds, if you try to catch all the clouds in one sim, this will be too big in size and too many empty area will be covered.

Separating the sim is a most likely option. But with that said, there is a portion of the sim I cannot separate; the long column flowing the rocket trailing upwards for good 500 frames. I've already asked Ivaylo once, but the issue I had run into was the trail must be from the single simulation to produce proper look over long period of time. I run the simulation and was able to get a low res sim with about 25 gb of ram. But to run a resim would require double or quadruple the amount depending on how long the sim is. My 86 gb machine easily runs out of ram on a sim like that.

So my question was, if it is at all possible to partition the sim across few machines with ram being separated too.

[Ivaylo Katev]

i have never tried this, but perhaps cascade connection can help in this case. first you avoid to catch the entire track in one grid (the empty space problem), and second and may be the more important - the lower parts pf the track are in fact almost static, no need to simulate them with the same dynamics as the top parts, which are still fast moving. you can animate the spf to decrease, or to use auto spf adjustment.
you can actually make only one simulation for a part of the track (for example 100m long), and to play it with time offset at different positions and different orientation to prevent the periodical view of the smoke.

[ivaylo.ivanov]

There is an example called phx_cascade_smoke.ma that shows the last cascade method I was experimenting with. If you have sufficient overlap between the simulators, the simulaton doesn't look so bad. It depends on the velocity inside the container. Rocket type movement will work. Mushroom type movement will most likely fail. Blending the simulators together is also tough and the seam is visible from some angles. Because of these issues, we haven't made automatic setup for this, and it's not easy to recreate. It's even not ported in 3ds Max.