Again, thank you ;)
Ok, I understand part of what you explain...
Juste one thing I don't understand totaly :
Instead of having a fixed resolution icosphere, why not start with a screen space set of vertices which would be generated based on the current view location?
Can you tell me more about what you mean ?
Don't know if it's the same process.. But in an earlier project, I was using a projective grid.. Some good results but lot artefacts and problems when camera was too close...
My proposal, have first pass only to tessellate, and the second, using Vertex Shader for computing noise, would not be too much slower I think.. ? And could resolve totaly the problem ? (with a much more detailed sphere as base model to avoid tessellator limits)