sharedMesh vs. sharedNode

here's one for ya: what's faster (to draw)?

1.

  • create a TriMesh (a cube)
  • create 1000 shared meshes based on that TriMesh
  • draw the shared meshes



    or



    2.
  • create a Quad
  • create 6 shared meshes based on that Quad
  • create a Node
  • attach the 6 shared meshes to the node (to make a cube)
  • create a 1000 shared nodes based on the previously created node
  • draw the shared nodes :smiley:

I think on modern graphics cards geometry-instancing is fastes, because batch-count is much more important than triangle-count.



I did not recognize a big speedup when I used shared meshes. Only memory consumption was reduced. But I'm sure someone else here knows a better answer. :slight_smile:

1.

2 just doesn't seem to make any sense - you're using 6 times more meshes to draw identical geometry.

#1 obviously.

Sharing meshes will only improve memory usage on CPU, unless you use VBO or displaylists, in that case memory usage on GPU will be reduced and performance improved.