The choice of engine decides how lights are traced in the scene.
LuxCoreRender offers the following engines (integrators):
- Path (CPU/OpenCL): Unidirectional pathtracer that casts rays from the camera. Samples the whole film progressively. Supports all AOVs.
- Tiled Path (CPU/OpenCL): Unidirectional pathtracer that is almost the same as Path, but uses a special sampler which iterates over the image in tiles (this leads to a slightly lower RAM usage). The OpenCL version adapts the number of tiles that are rendered at once to the performance of the compute devices (GPUs/CPUs), so the tile size does not matter a lot (if rendering performance is bad due to small tiles, the number of rendered tiles is increased automatically). When the last tile of a pass is reached, it is split among the compute devices. Supports all AOVs.
- Bidir (CPU only): Bidirectional pathtracer that casts rays from both camera and light sources. Samples the whole film progressively. Supports only a subset of AOVs: RGB, RGBA, ALPHA, DEPTH and SAMPLECOUNT. It is recommended to combine this engine with the Metropolis sampler.
The sampler decides how points on the film are chosen (in which area to fire more/less light rays).
LuxCoreRender offers the following samplers:
- Sobol: Random sampler with an improved noise pattern. Supports adaptive sampling to spend more samples on noisy areas of the image.
- Metropolis: Sampler that spends more samples on bright areas of the image, thus rendering caustics much better than the other samplers. The main disadvantage, especially when rendering on the GPU, is the higher RAM usage than the other samplers.
- Random: Simple random sampler. Supports adaptive sampling to spend more samples on noisy areas of the image.
The light strategy controls how much processing power is spent on each light in the scene.
All lights in the scene can be sampled with the same probability (uniform light strategy), but if some light sources are much brighter than others, the image will be more noisy because a lot of processing power is spent on weak lights that are not contributing much to the lighting.
The solution to this problem are the power and log power light strategies. They dedicate more processing power to lights that are brighter.
The following light strategies are available:
- Log Power (default)
The sampling probability of a light can also be influenced with the importance setting of the light source:
Importance: How much processing power to spend on this light source compared to other light sources. Used to scale the light importance computed by the light strategy. For instance, if you set a uniform light strategy, a light with a user importance of 2.0 will be sampled 2 times more often than one with 1.0. If you use a power light strategy, the user importance will be multiplied by the light power.