I have corrected the pressure equation I previously used in my solver, so that now it handles correctly density ratios. It can handle density ratio of about 100, being perfectly stable as far as a reasonable time step is used.
In this video, density ratio is 0.5.
I have implemented a mixability parameter, which influence surface tension between two fluids.
In this video, both fluids have a mixability factor of 50%, which means that a fluid is less attracted by another fluid than by itself (in order to keep a constant density at interface).
I definitely want to port to 3D. The only contrainst is user interface. I'm currently thinking about a minimalist yet powerful way to edit as it is possible in my 2D application.
~~~~~~~~~~~~~
| VIDEO LOG |
~~~~~~~~~~~~~
Number of particles : 96717
Domain size : 0.72m x 0.35m
Time step : 0.0003s
Time scale : 0.1x
Particle spacing : 0.001%
Interaction radius : 300%
Gravity : 9.81m
Display target : velocity
Simulated time : 6.00s
Simulation time : 1h 49m 58s
Video time : 1m 0s
Simulation factor : 0.009093
Frame per seconds : 3.64fps
Seconds per frame : 0.27spf
@ivanisavich This simulation took 1h49m tu compute. There's about 100,000 particles.
I've already made an implementation of the solver with OpenCL, in order to run on the GPU. But my implementation did not give the acceleration I was expected, despite my high end GPU. So I currently stick on CPU version, since is quite effective and code is really more convinient to read than GPU one.
VelRyphon 1 year ago
This is fantastic. What's the simulation time like?
As a total simulation ignoramus myself, I'm wondering....how hard would it be to get this running with CUDA to speed things up, if you went the 3d route? Most big 3d fluid dynamic packages are CPU only....you could really get your foot in the door competition-wise if you developed some kind of GPU-based 3d fluid solver!
ivanisavich 1 year ago