MentalRay 3.3

[dleon]

MentalRay 3.3 is now GPU based

[Everybody buy you Quadro 4000; it's worth 24 dual XEON boxes]

 

NAB 2004, Las Vegas, NV (April 19, 2004) mental images®, a global leader in the fast-growing market for high-end 3D visualization software, today announced the release of version 3.3 of mental ray®, its rendering software. mental ray is used both for cutting-edge visual effects in major motion pictures and for the computer-based design of industrial products. This software further blurs the line between reality and computer generated images, even when rendering the most difficult real-world features, including hair, fur and human skin.

 

mental images developed mental ray 3.3 in close cooperation with NVIDIA Corporation (Nasdaq: NVDA), to utilize the new line of NVIDIA QuadroFX 4000 boards and its Cg programming language. The resulting software takes full advantage of the unprecedented processing power and programmability of NVIDIA's hardware technology. For the first time, extremely complex visual effects can be fully executed on this graphics hardware, reducing rendering times from hours to minutes.

 

``NVIDIA is committed to working closely with mental images to ensure that the most cutting-edge software technology, like mental ray 3.3, works seamlessly with our hardware, and fully leverages the speed, precision and power of NVIDIA's new GPU architecture,'' said Jen-Hsun Huang, President and CEO of NVIDIA. ``We value our strategic partnership with mental images and look forward to continuing our successful working relationship as our two companies continue to push the boundaries between reality and what is possible to replicate on a computer.''

 

The mental ray technology is an integral part of the world's leading 3D content creation tools, such as Maya® from Alias Systems, 3ds max® from Discreet, a division of Autodesk, and SOFTIMAGE®|XSI® from Softimage Co., a subsidiary of Avid Technology, Inc. (Nasdaq: AVID). mental ray 3.3 is already included in SOFTIMAGE|XSI 4.0 and Maya 6.0, both released at NAB. These products take full advantage of the new feature set of mental ray 3.3 and leverage the performance enhancements, allowing for faster rendering both interactive and off-line.

 

``The deep integration of mental ray within SOFTIMAGE|XSI gives us a huge competitive advantage in the 3-D industry,'' said Marc Stevens, Director of R&D for Softimage. ``We have worked with mental images for over a decade to build 3-D animation tools that meet the needs of our customers, and the latest versions of XSI and mental ray significantly increase our customers' productivity and their ability to rapidly create the most realistic computer generated images.''

 

A number of key direct customers of mental images, including Buf Compagnie, DreamWorks, ESC Entertainment, and Industrial Light & Magic have also contributed substantially to the specification of the extended feature set of mental ray 3.3.

 

``The power of the human mind is largely a reflection of the three dimensionality of the world that we live in. At mental images we create 3D technology that makes it easy for people to fully explore their creative potential,'' said Rolf Herken, CEO and CTO of mental images. ``We are extremely excited to be working closely with established leaders in the 3D field like Alias, Discreet, NVIDIA and Softimage and will continue, as partners, to drive the transition to 3D computing.''

 

About mental ray 3.3

In addition to significant performance improvements, especially when rendering scenes that contain hair or fur geometry, additional features include the following:

 

Accelerated graphics hardware rendering: Exploitation of available graphics hardware for high quality rendering, including support for OpenGL, NVIDIA's Cg 1.2 higher level language, phenomena and shade trees where hardware shaders are available. Automatic combination of hardware and software generated elements. Support for displacement mapping, order independent transparency, hair geometry, volumes, and high quality anti-aliasing

Detail shadowmaps: Combination of shadowmap and ray traced shadow algorithms for more realistic shadows

Subsurface scattering: Powerful, physically correct simulation of subsurface scattering effects, according to actual local geometric and optic properties

State shaders: Tighter integration with animation packages and a more powerful interface between mental ray and front ends, resulting in better control of application-specific rendering effects and phenomena

Conditional rendering of scene files: For improved selection and control of processing scene contents

OpenEXR image file format: Support of the OpenEXR high dynamic range image file format developed at Industrial Light & Magic, for texture and image output

* * *

[dleon]

MentalRay 3.3 new facts emerge

 

Don't believe the hype, said Public Enemy. I guess they were right. Here are the latest revelations on MentalRay 3.3

 

(1) MentalRay is not fully ported to the GPU. Most of the core code is still CPU bound.

 

(2) All MentalRay functions can execute on the GPU *_in theory_*.

 

(3) In reality many (or even most) functions will continue to execute on CPU.

 

(4) MentalRay 3.3 will still be drastically faster than 3.2 even if you render exclusively on the CPU. It would appear that they are using all manner of clever mapping techniques (a-la VRAY) and manner of clever caching algorithms (a-la finalRender Stage-1) to whip the buns off the competition. Things have gone well beyond final gathering photon maps.

 

(5) MentalRay's use of the GPU is confined to the use of special .cg shaders.

 

(6) Whenever the names of a standard shader and a .cg shader match, the .cg shader takes priority, and runs the process on the GPU. Whenever there is no identically named .cg shader, the standard CPU shader routine runs.

 

(7) If 100% of your shaders are .cg shaders, then 100% of your process runs on the GPU and 0% runs on the CPU. Hence, it is possible for all MentalRay functions to execute on the GPU

 

(8) Practically speaking it is impossible for 100% of your process to run on the GPU. One of the reasons why MentalRay has not been well accepted in Hollywood lies in the fact that nobody knows how to write the old shaders much less the new .cg shaders. Any studio would have to undertake a major new .cg development project that would go on for some time before they could claim that 100% of their shaders are GPU based.

 

(9) Maya 6's implementation of MentalRay is almost as badly botched up as version 5's. The problems are far more profound than ever I imagined. It seems that the whole way Maya writes MI files is screwy compared to the way XSI writes MI files. This is primarily expressed as inefficient translation of expressions and extra (useless and redundant) garbage in the MI files. Not only does this jack-up render time but it can lead to massive time losses spent debugging scenes and manually tweaking (editing MI files) to get rid of artifacts and glitches found in the your frames.

 

(10) MentalImages is deeply concerned about Alias's bone-rotten implementation of MentalRay. This botched interface work is giving MentalImages a bad name all over Hollywood and the world. As a consequence, Berlin is taking more and more responsibility for writing more and more of Alias's core code for them. This is not something they have been contracted to do, not something they are being paid for, and probably something they shouldn't be doing, but they are doing it none the less. In a state of acute frustration, MentalImages is handing them .cpp files and saying "Here! Do it this way!"

 

(11) The reason why the Maya code was botched probably has everything to do with outsourcing. It seems that Alias is doing less and less work on Maya in Toronto Canada, and flicking more and more bits of the core hither-to-and-yon to be worked on by various teams around the world. Folks are hesitant to say where the particular outsource culprits responsibly for this botched work are in the world, but they are saying that the sudden appearance of Hungarian Verbs in the new MEL operators is quite curious indeed.

 

(12) SoftImage XSI continues to be the way the truth and the life as far as MentalRay is concerned. Nobody comes to the true MentalRay except through SoftImage XSI (right now).

 

(13) Discreet's implementation of MentalRay is pretty good. Just about everything in the core package renders well with MentalRay. There is some word of incompatibilities with plugins such as AfterBurn 3.1, but the core works great. Unfortunately, Discreet is hiding a great deal of parameters from us. It would appear that the XSI user has a smorgasbord of parameters that we don't have access to. Hopefully, this will change with the next release of 3ds Max.

 

(14) Sub-Surface scattering is new in 3.3. Nope, it wasn't actually in 3.2. Those who have used it say that it is very, very complex. Another smorgasbord of parameters to figure out. It was all done by some sort of bio-physicist who made sure that all the params and the entire math system is physically correct. Supposedly, it generates some sensational output. Make sure you use .cg shaders though, because this can jack your render time out of sight.

 

(15) The GPU raytraced shadow maps are supposed to be the real thing. Better than anything anyone has ever done yet. Crazy real. Enough to fool the best eye.

 

(16) How fast is this going to be? What is the final upshot in terms of speed? How fast are we going to be able to render now? The bottomline is that this question cannot be answered. It all depends on your use of .cg chaders. If you use a ton of them, you get a radical speed kick. If you don't use any, expect to render as fast or faster than you currently can with VRay or finalRender Stage-1

[Nick Nakadate]

Um... You might replace "final render stage 1" with Brazil r/s.

Does anyone even use FR anymore?

Oh, in case you missed it: http://www.splutterfish.com/sf/sf_gen_page.php3?printer=1&page=ArtistSpot/Orphanage/Orphanage

 

NN