waterFX / 6 cores i7 / incredibly!!!......slow ;)

[jay_vortex]

Alright here it goes:

I have this water effect i have to do, you can see it on the screen capture.

http://imageshack.us/photo/my-images...ureivaylo.jpg/

Im currently working on a 6 cores i7 4.2 Ghz and 24 gig of ram.

I've thrown the simulation overnight yesterday and this morning was just at 37th frame... :/

We dont really have a choice to put A LOT of cells because the emitters are small. I think it stabilize at around 70 millions cells with adaptive grid on.

Ivaylo, do you think i could send you the max file so you can have a look at the

scene and tell us if you think we did something wrong or if its pretty much the best

Phoenix can give us right now in terma of performance for this kind of project/effect?

Is it suppose to be workable at this kind of ''resolution''?

Thx a bunch.

Jay.

[Ivaylo Katev]

no problem, send me the scene. btw there is a performance indicator, it shows the cells per sec.

[jay_vortex]

Ok Ivaylo,

its sended. Thx a lot.

I knew about the performance indicator but i cant remember at what number it was and the grid is adaptive so i guess it went up to a lot when the grid was at 68 millions cells.

Thx!!

Jay.

[Ivaylo Katev]

well, the liquid setup looks good, the simulation gives me 0.7M per sec, the spf limit is 8, so the expected performance in the worst case is 8*70/0.7= 800 secs per frame. so, this means 8 h for 37 frames, it makes sense. the mt conservation in some cases may also lead to slower performance (it's system dependent).
i would suggest to use cascade connected simulators, this technique is very suitable for your case i think.

[jay_vortex]

Hmm well never tought of doing this calculation to predict the simulation time, duuh. Brilliant.

Thx Ivaylo, at least now i know im on the right track and that its normal it take that long.

Im gonna try it again with the cascade technique well see how it look at the junctions.

About this problem when using simulators in cascade, where you see the two fluids they're not ''continuous'', is it something that will be fixed in the next version?

Because its definitly a good way to work but kind of limited because of this bug.

Thx a lot!!!

Jay.

[Ivaylo Katev]

well, this calculation is for the worst case, don't forget that you have adaptive grid, and the spf is not always at the upper limit.
about the cascade technique - the problem is in the rendering, the fluids are quite continuous, but the vray shows the border between the materials.
whatever, the interface is almost invisible, in such a high resolution you have good chances to make it invisible i think.

[jay_vortex]

Alright, so i tried cascading the simulators. You can see on the screenshot the separation at render time even with high cells count.

http://imageshack.us/photo/my-images/714/cascadej.jpg/

And maybe im wrong but we see the separation in the viewport also, making me think the fluid is'nt continous? (i know you said it was :P ) What do you think?

*EDIT* And with motion blur the gap is a lot more pronounced making the thing worst.

[Ivaylo Katev]

the grids must be more overlapped, about 4 cells.
here is the result form the test scene

[jay_vortex]

So i did a lot of testing yesterday with the cascading technique and even with a lot of resolution the ''seam'' is too apparent, and like i said with motion blur its even worst because its bluring the seam, making it bigger.

Small, medium or big overlapping of the simulators does'nt really help, it just put the seam else where.

Two things i noticed:

1- It seem that after you simulated lets say two grids, there's an offset in the frames. You can see on the screenshot the difference when you offset by 1 frame with the ''read offset''. So basically the next simulators does'nt always seem to match the one before him. Maybe its me, but it really seem like it..

2- In any tests i did with multiple simulators in cascade, my source for the next simulator need to have less density then the first one to match the first simulator's fluid. Or else the second simulator's fluid is bigger then the first one. I wonder why.
Inject or brush does'nt change nothing for this.

http://imageshack.us/photo/my-images...cadesbugs.jpg/

Its really a great technique, we need to have this working since it allows us to do some effects that normally we could not do because of computer performances!

Anyways, i can send you the test scene, if you take a look at it in regards to my observations. Maybe.

Thx!!!

*EDIT* Added the screenshot.

Jay.

[Ivaylo Katev]

ok, just to be sure that you are using the technique correctly:
- the first simulation should be performed first, the second must be started after the first one is finished
- the discharge of the second source should be some very small value (for example 0.001), because the velocity is copied from the first liquid. this is very important!
you can see how it should look:

[jay_vortex]

Yes, i did those two things. I looked your test scene, made sure everything was ok. It all seem good.

Velocity is at 1, on the sources too.

I'll send you the file.

Thx.

Jay.