Jump to content

  • Log In with Google      Sign In   
  • Create Account

We're offering banner ads on our site from just $5!

1. Details HERE. 2. GDNet+ Subscriptions HERE. 3. Ad upload HERE.


#Actualchrisendymion

Posted 08 July 2013 - 03:02 AM

Thank you again for your help ;)
My goal here, is to have a fully GPU rendered map (spherical). Without "limit", LOD is managed by the tessellator, details by the noise loop (octaves and parameters).
So "in theory", you can go very close to the map without loosing quality & speed.
 
Because trying solutions is a lot of work, I'm searching the "probably" best way before testing.
 
You talked about texture... You mean, first pass, generating a texture, and after reading it in the second pass ?
It could be interesting.. 
 
What about :
 
First Pass
- Passing a precomputed spherical model to the GPU
- Using tessellator with camera position & direction for adding vertices (LOD)
- Streaming the buffer output from the geometry shader
 
OPTION A
Second Pass
- Using only the Compute Shader for computing noise and normals
(In this case, what about parallelism ? How to design the loop ?)
 
Third Pass
- Pixel shader for lightning
 
OPTION B
Second Pass
- Using Triangle_Adj as primitives
- Computing noise in vertex shader
- Computing normals in geometry shader
- Pixel shader for lightning
 
Or maybe, a totaly another solution ?
I don't care for the speed for medium-low hardware.. Only the quality and a good speed on high power GPU (GTX780).

#1chrisendymion

Posted 08 July 2013 - 12:17 AM

Thank you again for your help ;)
My goal here, is to have a fully GPU rendered map (spherical). Without "limit", LOD is managed by the tessellator, details by the noise loop (octaves and parameters).
So "in theory", you can go very close to the map without loosing quality & speed.
For that, I think that precomputed map is not the solution (even it's a lot more faster).
 
Because trying solutions is a lot of work, I'm searching the "probably" best way before testing.
 
What about :
 
First Pass
- Passing a precomputed spherical model to the GPU
- Using tessellator with camera position & direction for adding vertices (LOD)
- Streaming the buffer output from the geometry shader
 
OPTION A
Second Pass
- Using only the Compute Shader for computing noise and normals
(In this case, what about parallelism ? How to design the loop ?)
 
Third Pass
- Pixel shader for lightning
 
OPTION B
Second Pass
- Using Triangle_Adj as primitives
- Computing noise in vertex shader
- Computing normals in geometry shader
- Pixel shader for lightning
 
Or maybe, a totaly another solution ?
I don't care for the speed for medium-low hardware.. Only the quality and a good speed on high power GPU (GTX780).

PARTNERS