On Windows with 5.4.4
A rough dielectric sphere with an albedo of (0,0,1) shows banding artifacts similar to quantization errors:
- No Lumen, Nanite or TAA
- The banding is independent of the mesh resolution
- Happens with default lit surface material
- Occurs with any light strength
- GBuffer is set to use
High Precision Normals - The mesh has correct normals and is set to
Use High Precision Tangent Basis - Occurs with floating point render targets
- Unaffected by shader float precision
- Can be reproduced with a custom unlit shader that does Lambertian shading using the world normal and a light direction
- The bands are oriented towards the light direction and get narrower as the normal diverges from the light direction
- Happens with point lights and directional lights
- Is not affected by shadow settings
- Occurs with all albedos except maybe black but is most noticeable with blue
- On a white sphere, the faint bands tend towards yellow
- Occurs with the default post process pass as well as custom tonemapping
- Occurs in the linear data if tonemapping is bypassed completely
- A custom quantization dither applied before output gets rid of another type of banding but not this one
Why does this happen and how can it be fixed?