Taichi_elements (MPM solver for Blender)

Artwork Blender Tests
1

Here I will be sharing some of the tests I did with taichi_elements which is a multi-material continuum physics engine that works inside of Blender and is coded in taichi.

Currently the Blender implementation is still very experimental/a work in progress.
If you want to try it for yourself and don’t know how to install it you can check this out

I am doing this now because cycles got a point cloud geometry type which makes rendering millions of points a lot easier/possible.

My PC specs:
GPU = 2080ti (11GB of VRAM)
CPU = i9-9900k (64GB of RAM)

TEST 1
Simulation:

  • Sand and Water
  • 1.5M points
  • 240 frames (at 60fps)
  • 6 minutes bake time (CUDA)

Rendering:

  • 1920x1440
  • 20 seconds per frame (256 samples)
  • Peak memory 400MB

TEST 2
Simulation:

  • Elastic and Water
  • 2.65M points
  • 300 frames (at 24fps)
  • 26 minutes bake time (CUDA)

Rendering:

  • 1920x1440
  • 20 seconds per frame (256 samples)
  • Peak memory 520MB

TEST 3
Simulation:

  • Sand, Elastic and Water
  • 6.8M points
  • 300 frames (at 24fps)
  • 40 minutes bake time (CUDA)

Rendering:

  • 1920x1440
  • 20 seconds per frame (256 samples)
  • Peak memory 1020MB
20 Likes
2

I went and checked the activity of the repositories.

Taichi still looks active:

2022-01-02 13_15_09-Window
2022-01-02 13_15_09-Window926×648 36.2 KB

But taichi_elements looks a lot quieter:

2022-01-02 13_14_53-Window
2022-01-02 13_14_53-Window931×654 27.4 KB

I hope that this is just a temporary thing. Maybe there is some kind of a limitation from the Blender side of things that would limit future development? Or maybe the developers are just busy with other things :slight_smile:

3 Likes
3

those looks fabulous ! will there be a video to show how to use it inside of blender ?

3 Likes
4

Maybe in the future when it gets more developed but for now, you can look at this screenshot of the node system (this was for TEST 3) to get the idea of how things work:

taichi_nodes
taichi_nodes2709×1065 268 KB

5 Likes
5

Interesting…

6

Dynamics seem to be a big Blender hole.
I would have preferred that Nodes started with Dynamics instead of Geo, the impact would have been much bigger.

7

They tried to start with that but didn’t go anywhere (was too ambitious, too early), so they decided to be more “realistic” and go one step at a time starting with GN.

8

Well, you have to walk before you can run. You need geometry nodes to prepare emitters and colliders.
This kind of stuff

Currently, a big missing part of geometry nodes is in the UI for me. There needs to be a way to expose curves or color ramps so that you don’t have to open up the whole geometry nodes graph. Some kind of a visual separator between inputs would also be nice

4 Likes
9

Thanks did not know that. When i was in Softimage they started ICE with particles dynamics. I wonder it had implications for a more difficult implementation of mograph and geo stuff later.

10

Blender could really benefit from a system like this, I wonder if any of the devs hang out here?

3 Likes
11

@HooglyBoogly if you pass by take a look at the 2 first post

1 Like
12

@koko_ze
This looks sick dude. How does it compare to Manaflow?

Have you seen this person’s real time simulation engine? Any chance y’all could collab? They seem to need the help to further development. Best of both worlds?

1 Like
13

Sorry for the confusion but I’m not one of the developers of taichi/taichi_elements. I just play around with it. Yes, I’ve seen this and it does seem pretty cool.

Regarding the comparison between Mantaflow and Taichi_elements it is currently too early. Since Taichi_elements can run on the GPU it should be faster but it also doesn’t currently have a FLIP solver (instead of simulating all of the particles it could simulate them only at the borders) or mesh collisions which makes it a bit unpractical for anything serious at the moment.

14

Ahh. Gotcha. Good to know :slight_smile:
Thanks for playing around with it and giving us exposure to it!

1 Like