Optimized ray tracing

From Wiki

You are here: mental ray cookbook / optimized ray tracing

Contents 1 BSP 1.1 Large BSP 1.2 BSP2 2 External links

BSP

The BSP tree is an acceleration structure to speed up raytracing. Basically, it takes the entire scene space and recursively dices it into subspaces by computing dividing planes. When tracing a ray, mental ray then uses that tree hierarchy to quickly find out which triangles it needs to test for an intersection.

In order to do that effectively, the user needs to tell the system when to stop subdividing the scene by specifying a maximum tree depth (#1) and a minimum triangle count (#2) per tree leaf. These numbers balance rendering speed versus memory consumption.

Drawbacks:

• Does not always generate an optimum BSP (hence the "depth" and "width" parameters that the user have to put in and that are hard to tune)
• BSP does not care about instances, they will be converted to copies.
• Has problems with motion blur, where a fast moving object causes "long and thin" BSP leaves which are very inefficient.

Large BSP

The main advantage of the Large BSP is that it can be flushed, if memory is getting low. This means if you run into memory problems, Large BSP might help. On the downside is that it needs more memory to setup.

BSP2

Starting with 3ds Max 2009, mental ray now has a new type of BSP, called "BSP2" which is supposedly better at handling motion blur, instanced geometry and dynamic scenes. See [1] and [2] for technical background information on this new acceleration structure.

External links

• BSP Explained To Artists by Bernard Lebel. This article discusses the mental ray BSP tree and proposes a methodology to find very quickly the right values to use for BSP tree parameters. In essence, we'll look into how to use mental ray's diagnostic modes to make our render faster. But this article goes beyond the simple use of diagnostic modes, it also discusses the nature of the BSP tree.