These settings control Arnold's tessellation of subdivision surfaces. Note that, as well as the global subdivision control described below, you can also control the subdivision of an individual object, via the Arnold settings on the Attribute Editor. Select the geometry object in question in Maya, go to the Attribute Editor, scroll down and expand the Arnold group, and choose a suitable value for 'Subdivision Iterations' (there are also other controls for per-object subdivision. The actual number of subdivisions for each object will be the lower of the two values.
This value sets an upper limit on the number of subdivision iterations for all objects. By default, this is set to a very high value (999) which in practice has no effect. Setting this to a low value such as 1 or 2 can be useful when debugging scenes that take a long time to subdivide/tessellate.
Cube subdivided with 0, 1, 2 and 3 iterations (objects shaded with the utility shader in polywire mode).
Subdivision patches outside the view or dicing camera frustum will not be subdivided. This is useful for any extended surface that is only partially visible as only the directly visible part will be subdivided. Similarly, no subdivision work will happen if a mesh is not directly visible. This can be turned on globally by setting
options.subdiv_frustum_culling true and can be turned off for specific meshes with polymesh
Adds a world space padding to the frustum that can be increased as needed to minimize artifacts from out-of-view objects in cast shadows, reflections, etc. Note that motion blur is not yet taken into account and moving objects might require some additional padding.
The camera to use when determining subdivision levels of patches during adaptive subdivision. When enabled, the user provides a specific camera that will be used as a reference for all dicing (subdivision) calculations during Adaptive Subdivision (in other words, the tessellation of the object will not vary as the main camera is moved). This can be useful to fix objectionable flickering introduced by Adaptive Tessellation with certain moves of the main camera. If you set a static dicing camera, you will still get the benefits from Adaptive Subdivision (higher polygon detail closer to the dicing camera) while getting a tessellation that does not change from frame to frame. By default, this is disabled, and should only be used when necessary, and with a carefully chosen position for the dicing camera.
Polygon plane rendered with dicing camera (Catclark subdivision with 10 iterations)
Dicing Camera menu found under Subdivision in Arnold Renderer tab of Render Settings