Blender needs more 3d fluid engine

Hi

As you know the fluids go real bad when colliding 3d objects, or plane at angle other than vertical or horizontal.

Current engine is a little off balanced.
I can render in 0.04 x 0.04 x 0.04 m domain or in 2.7 x 2.7 x 0.15 m domain from resolution 300 - 500. Memory use goes to 7 Gb, one fluid frame mesh has 4 000 000 - 6 000 000 vertekses.
But the stair pattern will ruin all the work. The pattern will affect the opposite fluid layer surface when it’s thinn. It’s like the max fluid simulation Z resolution is 4.

6.5 Gb to render 4 step stair at resolution 450

Fluid revision is part of the new Salad branch (Carrot). So there is someone currently working on improvements, in theory.
http://www.graphicall.org/tmp/gsoc_salad.png

The salad branch looks like a christmas tree with gifts :slight_smile: I sure hope they all find their way in the trunk.
But as the GSoC guys are all known, I can make a journey and visit them asking why they didn´t finish their stuff lol.

That said. Buy more memory lol. Today it dropped to ~4.60Euro per GB.
DDR1300, 8Gib, 37 Euro.
Will be nice for Llano and Bulldozer as they like fast memory :slight_smile:

And now seriously, I didn´t follow Carrot Branch, but there was a project for OpenCL fluids in Blender somewhere around, I wouldn´t be surprised if Chrispther Neal is that guy. I am too lazy to research atm. I guess you just have to wait a bit until its running, or go and buy a tool for liquid simulations.

That’s cool someone is working on fluids.
When GPU’s are going to render the fluid’s is there enough memory?

How doe’s the current fluid’s system compare to RealFlow 's particle fluids.
I did test the RealFlow another day and it was like fluid was a lot balls and there was option to make shell around all the balls. I imagine it’s possible to feed % of balls to one GPU and some to other GPU or and CPU. It’s gonna be like 500x faster than 1 core of one cpu that may be years old c2d.

BTW Nice branch diagram :cool:

You should check out the carrot branch if you get a chance, or the whole salad! Check out these videos, the second one looks like he used a combination of the new dynamic paint waves and the fluid simulation.
http://www.miikahweb.com/en/blog/2011/07/11/testing-blender-camera-tracker

The dynamic paint waves simulation is pretty amazing, while not completely physically correct it’s still pretty useful.

I’m not so sure about the comparison between Realflows particles and the current implementation. I do know some time ago there was cool development going on in this area, haven’t seen anything recently though.
http://enja.org/2011/02/10/particles-in-bge-fluid-surface-and-collision-in-the-works/

Cool videos

I was wondering if the surface reconstruction ball pivoting of point cloud would be time-consuming to export fully on GPU.
It will get ride of at least 95% of the self intersecting faces in fluid mesh, but is very slow on single cpu core.

I want DDR2 memory 4GB models, but they cost more than ddr3 now.

8GB KIT LOW POWER KTM5780LP/8G
2 x 4 GB 667MHz = 186.90 €

Now i have 4 x 2GB 800MHz (my CPU fsb)
To go up from 8 to 16GB it’s ~374 €