Rendering Algorithms
- Introduced detail shadowmaps
that can multisample shadowmap pixels, and also call shadow shaders on shadow-casting
geometry to obtain and store transparent shadows.
- Shadowmaps can now specify a shadowmap bias to replace the standard
Woo trick, which stores the
halfway distance between the first two shadow-casting objects, with
a fixed distance. The bias can also be set in the options. Although
Woo shadowmaps are generally superior, they are not compatible with
shadowmap merging, which was
also added to mental ray 3.3 to support shadows cast from one
rendering pass on another. Note that mental ray 3.3 will read
shadowmap files created with older versions of mental ray, but
older versions cannot read the new 3.3 format.
- Shadowmaps also support extra fields supplied by a camera
attached to the light source.
- Uncompressed HDR (High Dynamic Range) images are now supported
in addition to compressed HDR images.
- Performance improvements: Light emitter shaders now work
several times faster. Raytracing motion-blurred hair geometry, and
hair rendering in general, is also several times faster.
Significantly improved final gathering performance for surfaces
with very strong bump mapping;
for extreme bump mapping, performance can increase by a factor of
10. Scenes with very large numbers of instances are rendered much
faster with far reduced memory overhead. Frame buffers now take
about half as much space.
- Large finely displaced objects of which only a very small part
is in the viewing frustum, and the bulk of the object to the side
or behind the camera, are now handled much more efficiently. Much
less subdivision effort is expended on invisible portions of such
objects.
- For hyperthreaded Intel CPUs and Linux, the -threads
command-line option now includes hyperthreaded pseudo CPUs. These
pseudo-CPUs will not consume licenses.
- Autovolume mode no longer requires ray tracing.
- Multihosted rendering could get wedged in mental ray 3.2 under
certain circumstances. mental ray 3.3 has a new networking
algorithm that is completely stable, and also much faster,
permitting more slave hosts to contribute efficiently.
- The multipass rendering
file format was changed. Old files can be imported, but older
versions of mental ray cannot read the new format.
- mental ray is now much more frugal with stack allocations. The
default stack size was reduced from 4 MB to 512 KB. This reduces
memory requirements and the risk of being unable to start another
thread. If 512 KB is insufficient, it can be overridden with the
MI_STACKSIZE environment
variable.
Copyright © 1986-2006 by mental images GmbH