Geprüft mit Version: 5.5
Schwierigkeitsgrad: Fortgeschrittene Anfänger
Up to this point we have discussed Precomputed Realtime GI in terms of Charts. Reducing or otherwise optimizing the number of Charts in our Scene limits the number of operations such as lightmap compositing and packing required during the precompute. This benefits precompute performance and is a broad approach to reducing the size of the data set required by Unity’s Precomputed Realtime GI solution.
Modifying Clusters is a finer-grained approach which will allow us to reduce the number of operations needed during later tasks in Unity’s precompute process. An additional benefit of reducing the number of Clusters is that run time performance will also be improved.
When generating the Scene’s lighting solution using Precomputed Realtime GI, Unity simplifies the calculations required by working on a voxelized ‘proxy’ of the Static Scene. These voxels are called Clusters. Clusters are effectively surface patches (small tiles) mapped onto Static geometry which we use for lighting. Clusters are stored in a hierarchical relationships and are used for the complex irradiance calculations needed when precomputing Unity’s diffuse global illumination solution. Note that although Clusters are mapped in a similar way to Charts, the two are actually independent.
The Clustering Scene draw mode can be used to visualize the size of clusters generated by Unity’s Precomputed Realtime GI.
Clusters sample the albedo of the Static geometry to which they are mapped. Then, during the Light Transport stage of the precompute, the relationship between these Clusters is calculated so that light can be propagated throughout the Cluster network. Unity generates this low resolution approximation of the Static Scene in order to simplify the amount of lighting data which needs to be updated during run time in order to deliver global illumination at interactive frame rates with the limitations of current hardware.
Illustration showing how the lighting value of Cluster X is related to the value of nearby Clusters (Image courtesy of Geomerics).
Once the precompute has been completed, ambient (skybox) lighting along with light positions, intensities and colors can then be modified without needing to restart the precompute process. These lighting changes will bounce and permeate throughout the Cluster network, taking into consideration the underlying albedo and emission of the Scene’s materials in the eventual output.
Following this initial bounce, updated lighting results can then be applied to the Clusters themselves. Following each iteration, lit Clusters will be sampled into the corresponding lightmap texture before finally being used by shaders within the Scene.
As this process is performed asynchronously on the CPU, the time it takes to refresh the global illumination solution is bound by the number of available worker threads. If needed, the amount of worker threads can be controlled using the ‘CPU Usage` setting found in the Lighting window.
The Lit Clustering Scene draw mode shows the Clusters once light has been bounced throughout the Cluster network.
Clusters can be visualized using the Clustering or Lit Clustering Scene draw mode. Using Clustering, the Scene will be overlaid with a diagnostic pattern made up of multicolored squares. The size of each square represents the size of a Cluster when mapped on to the corresponding Static geometry. Each unique color represents a different Cluster within the Scene. Similarly, the Lit Clustering Scene draw mode shows these Clusters once lighting has been bounced throughout the Cluster network and the results written back to the Clusters.
Reducing the number of Clusters largely determines how quickly this update occurs and, as a consequence, how interactive Unity’s Precomputed Realtime GI appears on your target platform. More importantly, for the purposes of this tutorial, reducing the number of clusters will improve the time it takes to precompute your lighting. As a result, iteration speeds when lighting your Scene will also improve. We will move onto techniques for reducing the number of Clusters in the following section of the tutorial.