Foam much slower to render in PhFD 3.12 than in PhoenixFD 3.0

[Zakkorn]

Today I realized that I was right when in a previous thread I made (about the ocean) that the foam in Phoenix FD 3.1+ is much slower than in Phoenix FD 3.0

I tied to make something with foam again.
But when it rendered it took an infinity to finish.

I tried then (because I had test it when I bought this machine) the Beer sample scene.
In frame 27 (I was bored to finish the whole sim) it takes 1min 27 secs in phoenix 3.12 to be rendered.
But in PhFD 3.0 it takes 45 sec

I cannot imagine how long will take in the final frames when the foam takes about 10% of the glass in size.

If you changed the algorithm for good, why is so slow?

Also in my project I noticed that id I render 1st time with foam calculations appeared almost instantly and the render begins.
But If I re-render any frame after that it will take about 15secs to begin the foam calculations.
Why is that?

PS. Now I noticed that if the foam is a lot (lets say close to a million particles) it takes about 1 (or more) min for transform vertices finish and the make the foam calculations.
I didn't have that kind of problems in the past.

[Svetlin.Nikolov]

Hey,

I just need a link to the exact scene you are testing with and the exact versions of V-Ray and Phoenix you are using and we'll track this down. If it has slowed down, we definitely have to fix it.

Cheers!

[Zakkorn]

As I said I came to this realization on the ocean test.
And I confirmed that with the Beer example scene from your tutorial section.

Did you made the foam slower to render when in bubbles?

Regards,
Zach

[georgi.zhekov]

Hey,

Are you using the exact same scene when testing both versions?

Can you share the scenes you are using?

What are the exact version of Phoenix and V-Ray you are testing with so we can investigate what is going on?

Thank you,

[Zakkorn]

I am using the same scene of the beer yes.
I just have to recalculate the sim (thats why I left it till the 27 frame)
And as I said to the first post I remembered when I bought this machine that was very fast to render that scene back then but now it seems pretty slow for some reason.
And then I reinstall PhFD 3.0 and test it and so that there is a big difference in foam render speed in compare with the latest versions.
I am using Vray 3.6.05

Regards,
Zach

[Svetlin.Nikolov]

What numbers are we talking about in render speed difference?

Also, is this the scene you are using? https://docs.chaosgroup.com/display/...#Examples-Beer

Thanks!

[Zakkorn]

In first post I gave the numbers.
Beer scene Frame 27
PhFD 3.0 45 secs
PhFD 3.1.2 87 secs

Regards,
Zach

[Svetlin.Nikolov]

Roger, got it finally! The difference occurred between Phoenix 3.03 and 3.04. Interestingly, we have that same scene as a test scene and we're measuring performance on it, but it renders using the progressive sampler and hasn't slowed down.

Will ping you when I know more...

Cheers!

[Zakkorn]

Does this has to do with the fact that I am using a Numa machine?
Because some bugs I found out in the past there were based on the Dual Xeon architecture.

Such as the slow scene transformation when I have a lot of foam.

The scene I have is t the old one so it has the bucket enabled.

But in progressive how do you know that is the same speed?

Regards,
Zach

[Svetlin.Nikolov]

Hey, found it.

There was a bug up until Phoenix 3.04, where in the Particle Shader the connected Liquid Geometry (now renamed to Liquid Simulator), was not working in some cases. The beer scene is exactly one such case - since Phoenix 3.04 the connected liquid simulator starts affecting the displacement and the tint of particles in the beer scene, but since you don't actually need these, it only slows down the scene needlessly. If you turn off the Liquid Simulator checkbox the speed will be back to normal. And indeed, changing the scene to progressive seems to not be the only change - for the test scene the liquid Geometry/Simulator box is also turned off and this is why there was no difference in speed.

Hope this helps!

[Svetlin.Nikolov]

Still gonna check if this can be automatically sped up though...

[Zakkorn]

Try to recreate what you said but I can't

If I turn off the Liquid Simulator and hit render it crashes max. Turning off = I am unchecking the box of Liquid Simulator in the Particle Shader.
Or I understand you wrong? (no idea)

And what is the Geometry/Simulator box that you turn off as well?
(I think I am blind or I have no idea what I am doing) :cry

Regards,
Zach

[Svetlin.Nikolov]

Yess, just got it here as well, great catch! Gonna be fixed in tomorrow's nightlies and I will also check if it can automatically render more quickly when you have the liquid simulator connected but it's not being used.

[Zakkorn]

OMG I found a bug, again?
Always I begin blaming my machine of how slow it is as it is a Dual Xeon thus, it has some compatibility issues because of the architecture, and at the end it was just a bug.

I am glad that I helped though

Regards,
Zach

[Zakkorn]

Now we found a bug lets go to the other issue of the foam
As I said when I have a lot of foam (close to a million and more) the first time I hit render the scene transformation takes a sec to finish and then enter foam calculation stage, render and so on.
But after that every time I hit render will take about 15-20 sec to finish just the scene transformation stage.
This reminded me back when I found that there was a bug with Dual Xeons setup when they had to render Displacement (I rerported this in 2015 sometime)
It was doing exactly the same issue.
When I had big displacement (back then) it took about 1 and half minute to finish the transformation stage (and it was random as well, sometimes took 40 secs sometime more than 1:30) on my Xeon machine. But then it rendered normal.
But when I tested the scene on a six core i7 the transormation took less than 5 secs.
So that was a bug for the Dual Xeon machines set up.

I have the same feeling for the foam issue, that is because of the Dual Xeon setup.
Can you check that?

Do you have any discord channel as I can explain things faster and better through voice than through text?


Regards,
Zach