- CUDA GPU host memory usage (missing estimation of memory used by integrator and kernels)
- Shadow catcher
- Volume rendering
- OpenVDB volumes
- Mesh interior volumes and volume stack
- AO pass
- Shader ray-tracing (bevel, AO node)
- Tiled rendering for very high res images (but not for performance tuning)
- Multi-device rendering
- Burley SSS?
- Light baking (add init_from_bake and use regular integrator kernels)
- Data baking (dedicated kernels)
- Cryptomatte postprocessing
- Save buffers
- CPU profiling stats
Please keep Burley, Random walk in it's current state is very finicky and not reliable for all looks. The color doesn't converge to the correct one at very small radii, it's hard to exclude parts from affecting others except by separating meshes as different objects (other render engines let you exclude parts from each other in the shader), the sharp edges are not always wanted, and so on and so forth.