PDA

View Full Version : Lightwave vs. Mental Ray



duke
07-13-2007, 07:59 PM
This is a limited test, like most tests, but I wanted to compare the two in terms of high poly count, and AO calculation times. I'm unsure of how Lightwave's AA translates to MR's AA so they're probably a poor comparison. I'm no MR noob so the settings are tweaked to the fastest I could get it, and as you can see, Lightwave "lays the smack down"!

Athlon X2 3800+, 2gb ram
929,814 tris
12 AO samples

Settings in both scenes in terms of camera angle, FOV, light position and strength are exactly the same.

Lightwave
No AA - 12.2 secs
http://dukecg.net/LW_TempleSpeedTest_noAA_12-2secs.jpg

4 Samples AA (Fixed) - 40.8secs - This is actually alot better than the MR render (and quicker too), but as I said I don't know what the translation is so the no AA renders are a better indication.
http://dukecg.net/LW_TempleSpeedTest_noAA_40-8secs.jpg


Mental Ray (XSI 6.01)
No AA - 32.75secs
http://dukecg.net/MR_TempleSpeedTest_noAA_32-75secs.jpg

1 AA sample (non adaptive) - 1:40 mins
http://dukecg.net/MR_TempleSpeedTest_1AA_1-40-94mins.jpg


Fprime
Stopped at lvl4 - 25.2secs
http://www.dukecg.net/FP_TempleSpeedTest_0001.jpg

Amurrell
07-13-2007, 08:58 PM
Nice to see the comparrison, and good news for LightWave.

duke
07-13-2007, 10:11 PM
Further settings info:

Lightwave
Perspective Camera
Box Filter, Fixed Sampling
Black backdrop colour
Default Occlusion shader in the node tree, plugged into color. 12 Samples, Infinite.
Ambient at 100% (MR doesn't have an intensity, only a color so it made sense to do this so it would be exactly the same as in MR).

Mental Ray/XSI
Primary Rays - Scanline
BSP: 35 max depth, 30 max leaf size
0.25 Mesh splitting
Box filter with 1.5 size

duke
07-13-2007, 10:17 PM
I also tried to do it in Modo but I can't emulate the same settings so it wouldn't be fair:

-AA doesn't AA Global Illum like these 2 do.
-You can't simply apply ambient occlusion to the color, theres only an ambient occlusion rendermode which then excludes rendering the main light+shadows.
-For my purposes it would be interesting but fairly useless - i'm comparing programs that can animate!

t4d
07-14-2007, 01:01 AM
Mental Ray is not a Let's jump in and do what i do in LW type render

It's far more advanced then that, and when you do REAL testing will always show why it's a render guru's favour tool
( well one of them Pixar or 3Deight are there too )

but we are all not render guru's working with the job of rendering all day long so LW has it's place for it's ease of use,. but that's it.

But to compaire the 2 in a Serouis way sorry LW won't come off too well.

http://forums.cgsociety.org/showthread.php?t=389990&page=12&pp=15

Look at what Saturn does with the scene and check his times out here..

I uses LW as my render for the last 2 year with XSI or Messiah as the animation tools, But now i have really started to uses Mental Ray it's hard to learn but has ALOT of room to improve speed in ALOT of area's

it would be interesting to maybe start a thread and Drop the scene here on the forum and Maybe find a Modo guy and a Vray guy and we all do the same as that CGtalk forum and see what we can push out of each render ?

Cageman
07-14-2007, 01:41 AM
But to compaire the 2 in a Serouis way sorry LW won't come off too well.

Until you turn on Motionblur. Why do you think Zoic used LW for rendering the VFX in Drive if it didn't improve on what they had in Maya? Why take the extra step into LW if the result wouldn't be as good or better? Their pipeline for Drive was based on Modo->Maya->LightWave (Modeling->Animation->Rendering). So, with what you say, the last step wouln't have been necessary. I know MR can do alot of stuff and I really dig the fast, good looking SSS (misss), but bringing things out on schedule, esp. using MBlur is a HUGE challenge, and that is where MR falls behind... :/ I can get things out through LW with "real" 3D Mblur faster than MR can using Linear (only translation, no deformation) Mblur without loosing quality.

t4d
07-14-2007, 01:48 AM
Until you turn on Motionblur. Why do you think Zoic used LW for rendering the VFX in Drive if it didn't improve on what they had in Maya? Why take the extra step into LW if the result wouldn't be as good or better? Their pipeline for Drive was based on Modo->Maya->LightWave (Modeling->Animation->Rendering). So, with what you say, the last step wouln't have been necessary. I know MR can do alot of stuff and I really dig the fast, good looking SSS (misss), but bringing things out on schedule, esp. using MBlur is a HUGE challenge, and that is where MR falls behind... :/ I can get things out through LW with "real" 3D Mblur faster than MR can using Linear (only translation, no deformation) Mblur without loosing quality.


yeah and Drive was the only 3D production in the last 12 months ???
and Zoic is the only CG studio ?

Hey I'm first to admit Motion blur in MR Sucks !!
But with a good pass system it's not a huge issues to do it in post in most cases. BUT YES MR motion blur Sucks

but did you check that link out ?

duke
07-14-2007, 02:19 AM
T4D, my original tests were to test high poly count and AO. MR is probably as fast at AO, but Lightwave sure seems to chew through the polys. MR will go a hell of alot further with polycount than Lightwave because it has more settings for memory optimizations, where Lightwave will just crap out once it reaches a certain point, but again my tests were aimed at 2 specific parameters - high polys in the range of 1mil or under, and AO.

I'm familiar with that scene/thread you linked, i was F5'ing it once an hour :p when it started, but it's aimed at testing something very different. It's testing glossy rays (Lightwave is not the best at this and doesn't have any decent "tricks" to make it faster like MR and Vray do) and AO and FG/GI. I don't really care about glossy rays or FG/GI.

I'm not able to share this model around, and I don't think it would do any good anyway. Everyone has a different machine and even if we work out some retarded math to normalize the results it would become messy!

chunderburger (lovely) brings up a good point and I forgot to test it. What i'll do is apply different shaders to each column that we see, where they would be ones I can see myself using (glossy rays and SSS aren't one of them unfortunately), including high res textures on some of them and i'll see how it goes!

Cageman
07-14-2007, 02:33 AM
yeah and Drive was the only 3D production in the last 12 months ???
and Zoic is the only CG studio ?


Nopes, there are plenty of smaller shops that use LW, Zoic is just one of them. But they seem to have a very good understanding of for what and when to take advantage of a package like LW.

I wouldn't say that MRs motionblur sucks. It looks exactly as it is supposed to do, but it is darn too slow... :/



but did you check that link out ?

Yep and I noticed there were no MBlur going on and that's why I posted what I posted. :) I would also say that a chrome room with alot of chrome spheres aren't exactly "production". It is a great way to stresstest different lighting and shading techniques, which isn't a problem for any of the renderers today (getting good looking shaders and lighting). However, there are other things to concider when choosing a renderer...

I have been baybysitting alot of renders that doesn't render on the farm the last week just because of those other problems with MR, which are more related to memoryhandling and, probably, strange things that Maya has created in the scene that noone has discovered yet. And, the fact that we really need to render everything with MBlur. Adjusting the BSP settings has helped, but slowed the renders to even lower levels, but at least they do not crash. The fact is that, MR today (MR for Maya, I should say) isn't reliable as a renderer for production for a smaller team that can not assign one or more people to ONLY work as renderwrangler. MRStand-alone is probably alot more failsafe, and I bet MR for XSI is alot more failsafe as well.

There are other techniques to render mblur but they are all limited to only work in 2D rather than 3D (Motionvectors, camera projection etc). Works good for alot of shots, but anything moving in depth (related to camera, of course) will be hard to mblur using any of the pass-techniques. But for the shots that can use those techniques, ANY render can output images much, much faster. But it is always prefered to render as much stuff as possible "in camera", and mblur is one of those, which brings me to Renderman. It has the fastest mblur ever! Combine that with Displacement and memoryhandling and we have the most important reasons why Renderman is so heavily used in highend VFX-productions these days.

archijam
07-14-2007, 02:44 AM
t4D you make a good point, a comparison render of the same scene is a good idea, but I'm glad Cageman pointed out the motionblur issues .. you do sometimes neglect to mention if ever LW has strong points in your comparisons T4D ;) ...

.. a bit like saying 'modo is great for animation, you just need some post work' ...

j.

AbnRanger
07-14-2007, 02:48 AM
I have come to hate MR because it's so slow, and quite fickle, in my experience. The Distributed Render feature works in limited situations, etc.
Personally, I wish Max had either finalRender or VRay integrated and kicked MR to the curb. I use finalRender Stage 1, and it is SO much more compatible with Max plugins, and integrated than MR...plus it's Ssssooooooo much faster. It just outclasses MR in every category.

There's no way Mental Ray is as fast as LW's renderer in 9.2...not sure why you are supporting MR based on speed comparisons when this has long been their Achille's heel

t4d
07-14-2007, 02:54 AM
Well I find LW better at Cheap render times ( whole simple scenes where every pixel need attention But nothing too heavy to cal out )

it's hard to get MR to render a frame under 30 -45 second when every pixel needs attention where LW can do it in 5 -15 seconds.

I also Find LW better a motion blur ;) and when your looking for a UNreal look LW really is great to explore, it just looks cool and you just enhance it

Where MR can look bad and you have to play and learn alot to get what you want. But once you learn the basic's things move alot quicker.

MR if your pushing cals be it AO, SSS, GI, FG etc has Alot more room to play with how the system works it all out and renders it,
be it in the Shader/render trees, Passes and render settings and MI direct render settings,.

So if you have 1000 frames ahead of you, in MR you can bring render time down from 45 mins a frame to 5 - 10 mins a frame, so a day in R&D can save Days in render times,

BUT with that if you have afew CPU's around you can just get LW to throw more cpu's at the problem and get the same result. =)

I like LW for low polygon stuff BUT I'm finding more and more reason to do it all in XSI with mental ray then Export to LW. I have to say Mainly due to the pass system, I really have fallen for rendering in Passes it so powerful and saves So much re-rendering. also Displacement Render Very quick in MR And I seem to add those to everything nowadays . ;)

for me Mental Ray biggest negative is also it's positive...
It's so huge it takes alot to learn But all those options can be Very cool once learnt.

Cageman
07-14-2007, 02:55 AM
It just outclasses MR in every category.

finalRender doesn't have a "photoreal" mblur, like LW9.2, MR and Renderman and the tests I did showed that finalRender had a slower mblur compared to MR (and I used MR Exact setting). :)

AbnRanger
07-14-2007, 03:02 AM
are you talking all MR's or maya MR or MAX or XSI ?
Max.

AbnRanger
07-14-2007, 03:07 AM
finalRender doesn't have a "photoreal" mblur, like LW9.2, MR and Renderman and the tests I did showed that finalRender had a slower mblur compared to MR (and I used MR Exact setting). :)
Are you comparing the newest upgrade of finalRender...Stage1 R2 (just came out a few months ago)?
I believe they optimized the 3D motion blur in this version, plus motion blur for maps.
http://www.finalrender.com/products/feature.php?UD=10-7888-35-788&PID=36&FID=359

I'm with you on LW's 9.2 improvements.

I use Distributed Rendering alot...mostly for previewing purposes, and finalRender's DR is leagues better than Mental Ray. No comparison at all, really

oDDity
07-14-2007, 03:14 AM
You guys seem to be only talking about simplistic arch viz type situaitons. There are a lot of other scenarios where the customization possibilities of mental ray makes it far superior to LW, not to mention the superior shaders and the fact it is integrated into most major apps these days, whereas with the LW renderer, you're stuck with using layout, complete with all it's limitations, and who really wants to be bothered porting everything over to LW at render time. For complex scenes, it's not even possible.

Lightwolf
07-14-2007, 03:25 AM
There are a lot of other scenarios where the customization possibilities of mental ray makes it far superior to LW, not to mention the superior shaders and the fact it is integrated into most major apps these days, ...
However, the qualiity of the integration varies greatly. Try using memory mapped images with max (a mr feature) - no go.
Which also means: The learning curve includes finding out how the renderer works in the first place, and what part of it does or does not work with your package (something FPrime or KRay users are familiar with).

Cheers,
Mike

Cageman
07-14-2007, 03:26 AM
...and the fact it is integrated into most major apps these days

Integrated is hardly the word I would use for the Maya-version. As for the rest, I agree, except the things I've already mentioned (Mblur, stability issues etc)...I've also seen that there are a bunch of people creating stuff in LW that looks really, really good, so I'm saying MR and LW are on par with eachother in most cases.

Ohh...I'm not talking about archviz here... just go to www.worldinconflict.com and click on trailer to see what I'm working with. :)

Cageman
07-14-2007, 03:43 AM
Are you comparing the newest upgrade of finalRender...Stage1 R2 (just came out a few months ago)?
I believe they optimized the 3D motion blur in this version, plus motion blur for maps.
http://www.finalrender.com/products/feature.php?UD=10-7888-35-788&PID=36&FID=359


finalRender Stage-2 for Maya 7 and it was a quick test. But looking at the alpha channel I could tell that the mblur only accoured on one side of the object, the "tail". MR and LW9.2 applies equal amount of mblur on both sides, which generates a very similar effect to what a real camera produces; hence the "Photoreal" name in LW9.2.

Having said that, I don't think anyone would notice if everything was rendered with finalRender. Heck, I havn't seen much of the mblur banding artifacts that the Classic Camera produce, and it has been used in alot of movies and tv-series throughout the years... :)

Cageman
07-14-2007, 04:02 AM
.... but actually it is less expensive to buy the Mental-Ray licenses and have less work to do

I really hope Autodesk has fixed all issues regarding renderlayers in Maya 8.5. They are insanely unusefull in Maya 7.01 and has caused alot more problems than anything else. Since we are doing a huge production, we are more or less dependent on the Reference system, which also has issues, especially in combination with renderlayers...

The grass is greener.... or maybe not as often as we think?

Lightwolf
07-14-2007, 04:03 AM
... I dont understand why Newtek didnt see the power of an render-layer-passes-system...
Actually, I'm quite sure they see the power of a lot of things out there... but that doesn't mean it is easy to implement using the current base.
With limited ressources it probably makes a lot more sense to re-work the base before adding layers on top of them, instead of adding passes now and then needing to redo them once the base changes.

Cheers,
Mike

Cageman
07-14-2007, 06:32 AM
in maya 7 the render-layer system was new, they wanted to make it easy as the xsi system, 8.5 is better, but even with something like the old layer-system that was in maya since 1.0 it was less work, than save every object and scene separatly like in lightwave !!

Depending on what you want to do, you can use Renderbuffers in LW and only do scene-specific changes. Yes, you end up with alot of scenes, but you do not need to save out new objects.

Cageman
07-14-2007, 12:08 PM
... in some cases the render-buffer are not antialised, and if we need an occlusion-pass or a special shader-pass we have to save the objects

Have you tried TMP Save Objects? It saves objects enmasse with prefix/suffix and new path. For passes like Occlusion and other special needs, I've saved a ton of work using it. Highly recommended to ANYONE working with multipass rendering in LW (until NT implements a good solution for multipass rendering).

EDIT: Link to flay (http://www.flay.com/GetDetail.cfm?ID=1091)

Cageman
07-14-2007, 05:28 PM
... thank you cageman, it sounds great, I try a test next week

Here is a video (http://hangar18.campus.luth.se/~cageman/lightwave/TMPSave_1200x900.mov) showing it in action... no sound. :)

Stooch
07-15-2007, 01:48 AM
mental ray is a powerful renderer that isnt exactly the fastest but high quality and feature rich. The motion blur is a big deal for me and MR is very slow at that. I dont think its complexity is a negative but i do think that LW is competitive in smaller productions and with faster turn around. The new motion blur should give LW a really nice advantage that is bound to get some shots but it need comprehensive layering very badly. in general layout needs layers just like maya and even modeler. scene management is a huge pita right now in LW.

id love to see some really intuitive presets for render layers that generate all useful passes with one click. so load scene, assign objects/lights to render layer, make it renderable and choose the type of render passes you want from it. hit render scene and watch as lw creates all the passes, named accordingly to the layer and sorted into folders. maybe even use speed edit to quickly comp the passes in render view to see the entire shot. etc

also another major advantage is that MR is better built for distributed rendering. out of the box it can control 8 satellite cpus as long as it senses their presence.

Chris S. (Fez)
07-15-2007, 02:36 AM
id love to see some really intuitive presets for render layers that generate all useful passes with one click. so load scene, assign objects/lights to render layer, make it renderable and choose the type of render passes you want from it. hit render scene and watch as lw creates all the passes, named accordingly to the layer and sorted into folders. maybe even use speed edit to quickly comp the passes in render view to see the entire shot. etc


Amen.

duke
07-15-2007, 03:58 AM
setting up a nodal network on a pass would be sweet too.
so you could chuck pomm occlusion shader or whatever and render that out as a pass on everyting. none of your surfacing is changed and no assigning occlusion shaders to every surface of evey object.

actually, i put a feat req in for global scene and object nodes. this would be another, pass nodes

It's fairly late in the game, but i'm going to be moving back to LW for this particular project and i think i'll use POM's Store and Get Material nodes to do this kind of stuff. Sortof.

I'll have a box that has a surface where I have any global nodes I want to setup in it, then have all my store materials in there, and then for all my surfaces, anything I want to control from one thing i'll use Get Mat so all I then need to do is change the master one on that box surface. If that makes sense. It doesn't exactly address what you're talking about but those POM nodes sure will make a huge difference.

duke
07-16-2007, 06:32 AM
I'm not exactly sure. I don't know if they just pull defaults (and override the ones you have control over like super-sampling) or use your global settings. I suspect if you can make a rendertime difference and that difference carries over into ultimapper/rendermap, then it's using the render settings.

t4d
07-16-2007, 06:38 AM
for a job I just finished I was rendering over 20 million polygons
on a XSI 64 on Win 64 tho

and my settting were something like this
Mental Ray/XSI
Primary Rays - Scanline
BSP: 60 max depth, 25 max leaf size ( these setting I change alittle up and down depending on how much was on screen )
1 Mesh splitting ( I had ALOT of object Not one 20 million poly object )
Box filter with 1.5 size