Translated to english with Google Translate
In this paper, collection of tips for optimizing heavy scene to render in mental ray. Some of the tips apply to all cases, while others depend on the situation. Let's start in order of importance and effectiveness of the methods.
x64To work to only use x64 bit system and software. There is no reason to use x86 (x32). Time x86 systems has long passed. By the way, 32-bit software works fine on any 64 bit system. And it works, usually better.
FG distance, Occlusion and etcIf you have the exterior, you definitely need to limit the distance Final Gather'a, Occlusion, and that of similar effects. This must be done even if the scene you are not very heavy. This can seriously speed up render FG and significantly reduce the amount of memory. And the picture is almost not affected.
A typical example. There is a scene from the forest. To illuminate what is the FG point will emit rays in all directions over the hemisphere, loading all that came to geometry. In the forest, it will lead to what will be loaded all the trees around. But why do we need light came from the other end of the forest, when it will be virtually invisible. So it makes sense to take into account light from only near objects lying around the illuminated point.
For Final Gather this is the options 'FG Falloff Stop' and 'FG Falloff Start'. The main one is 'FG Falloff Stop', it was he who cut unnecessary geometry. A 'FG Falloff Start' makes pruning more gradual and less noticeable. As a rule the distance 'FG Falloff Stop' is calculated on the eye. Take the typical largest object in the scene, such as buildings or trees. And we put in 'FG Falloff Stop' 1-3 of height value of it. 'FG Falloff Start' put a little less 'FG Falloff Stop'.
For example, if the scene is the largest object is a tree height of 5 units, then 'FG Falloff Start' and 'FG Falloff Stop' you need to put a value of ~9 and ~12 respectively.
Keep in mind that when the geometry is clipped the sky (environment), which will illiminate on your objects. Therefore you should not greatly reduce the distance cut-off, otherwise you can get excessive illumination from the sky on all sides.
For interior use the falloff is not necessary, because the light should come from one wall to another.
BSP2If the scene is heavy and requires large amounts of memory, it is best to use the BSP2 acceleration mode. BSP2 an automatic mode with no options. Effective for heavy scenes, but for light scenes fit in memory Regular BSP is often faster.
Instance and ProxyIf the scene have a lot of copies of same object, 'instances' should be used wherever possible. Instances are loaded into memory only once, other copies is "free". Proxy is also very helpful, they are loaded/unloaded as necessary. This saves memory, but may not be very fast, especially in the case of FG and a massive raytracing. Therefore, the Proxy is better to use for unique objects, or when the copies are not so much. Or when a lot of unique objects so much that they all together will not fit in memory. In mental ray 3.8.x and above Proxy already working almost as instances. Ie, if using Proxy loaded many copies of same object, the memory is consumed as one.
Scanline | RaytraceFor primary rays is better to switch mode to 'Raytrace' (scanline = off). Often this can speed up the rendering a bit, as well as reduce the amount of memory. In some cases, particularly with heavy scenes, the memory savings can be significant.
NURBs, Displace and SubdiveIf the scene is NURBs, Displace or Subdive, then do not expect that their tessellation settings by default is always the optimal. In most cases it is better to configure it manually using Aprroximation nodes on the object. This is especially true when such objects are many and they are not fully visible in the frame.
MAP texturesIf the scene contain many textures, it's worth the largest of them to convert to the format *. map. This must be done so that they are not loaded into memory entirely, but only pieces as needed. If the texture resolution 2048 pixels or more, then we can think about converting to *. map. If size is smaller, it's likely to do so not worth it. Since the memory in this case saved the little. In theory, load the texture is often a little slower, especially on the network. Therefore, use the *. map just like that, without special needs, it is not necessary.
Conversion performed using utility imf_copy.exe. The command for conversion looks like this:
imf_copy -p -r texture.tif texture.map
And the flags "-p-r" are required, otherwise the texture will convert, but will still be loaded into memory entirely.
TreesIf you render a lot of trees, then the extra few tips:
If possible it is better not to use the transparency on the leaves. This can slow down rendering and increased memory consumption. Often, if the leaf shape is visible in the render, it is better to make the leaf geometry (without fanaticism). A long-distance plans fit a 1-2 triangle leaf.
No need to just include a reflection on the leaves and the etc. This is all strongly affect the speed and memory consumption.
Sometimes when the animation at the crown of trees can be placed lower settings for FG. Usually, there is no visible spot or noise.
And of course, do not use very detailed trees if it is not visible on render, particularly in the background. Realism, this does not add, but the problem...
Send your questions, and suggestions to firstname.lastname@example.org