A number of new features have been announced for the NVIDIA ARC release of mental ray 126.96.36.199 and saycon2000 has referenced them in a piece posted in the forums. Below is the cut/paste of the material. Makes interesting reading for those working in mental ray at present.
The Ptex texture mapping system and file format developed by Walt Disney Animation Studios is now supported for rendering of ccmesh subdivision surface geometry. The new base shader mib_ptex_lookup is provided that performs lookup of the texturing information at render time. Since it is utilizing third-party runtime implementation delivered with mental ray the shader comes as a separate package ptex.dll/.so/.dylib.
In order to enable Ptex per-face assignment the ccmesh surfaces need to carry additional data when computed. This can be enabled globally using a registry key, or per object with a new optional bary attribute.
The iray version 2.0 is a major update of this rendering mode built into mental ray. It provides numerous improvements in performance and quality, like native support for BSDF layering in the rendering core. This version is using the latest release of CUDA, thus it is supporting, and will actually take benefit from, most recent GPU hardware, like the NVIDIA Fermi architecture.
-Hair Ray Tracing Performance
Hair rendering with ray tracing uses a new acceleration structure, which results in significantly better performance and lower memory consumption. For backwards compatibility, the previous behavior can be enabled with a string option. The improvement is especially noticeable when computing shadows and detail shadow maps(10-30%). The image quality is increased at the same time by raising the intersection precision. In addition, a transparency depth limit of 250 has been implemented for hair ray tracing, similar to the the value used in the rasterizer, to avoid excessive computations for big chunks of hair.
-Detail Shadow Map Features
The quality of detail shadow maps has been improved. The data size of high-depth detail shadow maps has been reduced. Furthermore, a new string option allows to optimize size and quality of a detail shadow map.
-Unified Sampling Features
The support for unified sampling in mental ray and in shaders has been extended. The min/max sampling limits can now be specified per object. For convenience, the unified sampling feature can be controlled from the command line of a standalone mental ray.
-TIFF Image Format Features
The support for TIFF image format has been extended with more features, like
reading LZW compressed images,
reading and writing in tiled format, suitable for texture caching.
The output of tiled images in mental ray can be enforced with a frame buffer attribute. Using the imf_copy tool, a tiled TIFF file can be created using the -p option. Note, that the resulting TIFF subformat is not compatible with mental ray 3.8 and earlier, including its image tools. Tiled TIFF images may not be supported by external image manipulation tools.
-OpenEXR Stereo Support
The OpenEXR Multiview extension is now supported and used to create stereoscopic ("sxr") image files. This allows to store the images for left and right eye in a single .exr (or .sxr) file. This behavior can be enabled per frame buffer by specifying a datatype "sxr", or globally by using a registry setting. Note, that the imf_disp tool is able to display only mental ray generated multiview OpenEXR files. For other ones created with external tools, camera names may be interpreted as conventional layer names.
-Texture Caching Features
This version improves control of texture caching as well as reporting of cache performance and statistics. mental ray now prints out texture cache mode and cache size limit in the options info banner before each frame rendering. For each cached image file, ie. cached texture or frame buffer, it prints statistics about the number of pixel accesses and per-thread local cache coherency rate. At end of rendering, mental ray prints accumulated statistics. The dynamically computed texture cache memory is based on the size of the uncompressed textures in memory instead of the file size as used in earlier versions. In particular, the memory used for rendering no longer depends on the type of texture compression used.
-Subdivision Mesh Features
The ccmesh subdivision mesh geometry now supports approximation with the sharp property, to achieve faceted rendering look. In addition, per-face user data can now be attached to a ccmesh object. Finally, the previous limit of two texture spaces for ccmesh geometry has been lifted, and up to four texture spaces are now supported for texture seam computation.
-Initial Alembic Support
The Alembic open source interchange framework and file format is supported in an initial form. A new shader abcimport is provided to read static or animated geometry like polygon or subdivision meshes and NURBS surfaces from Alembic files, including support for loading them on demand only during rendering.
-Deprecated Scanline Features
This mental ray version deprecates scanline as its default rendering mode, but promotes ray tracing as the primary renderer instead. This allows to take benefit from greatly improved ray tracing speed using latest achievements in BSP2 acceleration, as well as noticeably reduced memory consumption especially for larger scenes by avoiding to keep extra rasterization data. To help transitioning, mental ray adds support for previous scanline-only functionality in ray tracing mode, like
clip rendering of geometry on camera hither/yon planes, enabled with a string option,
compute motion vectors including camera movement, enabled with a string option.
-New Shader Package
The new shader package useribl comes with mental ray. It supports convenient workflows for lighting from environment maps or from textured area lights similar to the built-in IBL functionality, this time provided as separate shader nodes with extended functionality.
-New Subsurface Shaders
Improved subsurface scattering shaders misss_fast_shader2(_x) have been added with extended features towards production purposes. Consequently, two skin phenomena misss_phen_skin2_mia_phen(_d) are available that are based on these shaders.