...
- Ray-tracing performance has been improved to such a degree that we now recommend using the ray-tracing hider as the default renderer for scenes with highly complex geometry (instead of REYES). Users can expect gains up to 20x faster when using multi-bounce global illumination and glossy reflections.
- Ray-tracing of motion-blur, even in the most extreme cases, is not a problem anymore. Performance is such that there are no objectionable speed differences between motion-blurred and static frames.
- Ray-tracing of dense and translucent surfaces, such as hair, is up to 3x faster.
- Ray-tracer startup time has been diminished. Space partition build has been multi-threaded and other, geometry specif optimizations, have been implemented.
- Sampling in the ray tracing algorithm has been greatly improved. This means that less samples are needed to obtain noise-free images, especially with image based lighting.
- Subsurface scattering is 5x faster when combined with the path tracer.
- Multi-threading scalability have been improved for both the REYES rendering algorithm as well as in the path tracer.
- Better multi-threading of texture reads. This can drastically improve render times when using a very large amount of texture files. We measured improvements of up to 25% faster.
- Rendering dense volume clouds (region with many interior shaders) is up to 2.5x faster.
- Point-cloud writing is now multi-threades.
- Shader compiler produces better code and is able to use up to 50% less varying variables as before. This translate to faster renders.
- Access to very large files have been improved by using memory mapped I/O when possible. This is especially useful when accessing large point-based files.
- Memory usage usage for very large polygonal meshes. We measured up to 4x less memory used on some typical production scenes.
tdlmake is 2x faster.
...