Maxwell vs Mental Ray animation test.

[mister3d]

But I want to tell you that imho there are obvious differences in setups:
the chrome material is much more glossy in maxwell
the shadows in mental ray look very dark, but I think you could use right exposure in mental ray to avoid this
the red material on spheres in maxwell is 2-layered like a car-paint(there are sharp reflections from square light and glossy from shining light material). This is for sure, and in your mental ray scene it is 1-layered material with fake highlights(no square light reflection) and no glossy reflection from shining light on the ball
it seems to me that in maxwell the light sources are colored, and in mental ray not. Maybe it's not so in your setup, but I had to color light sources in vray to achieve the result more similar to maxwell. The same with red color - shifted to green
the upper light is stronger in mental ray setup
maybe I'm wrong, but there's a vignetting turned on in maxwell?
It seems like the white box material in mental ray is more reflective in mental ray, I can see a certain reflection in the mental way setup

Did you use 1 machine or more? - "The times for the entire animation varied since our Farm is not same same for each node. So some machines took much longer per frame. But this gives you an idea."

Here are my quick results with vray. You can see the rendertime in render below

http://img522.imageshack.us/img522/8110/ballanimationvraynewtg6.jpg (http://imageshack.us)
http://img522.imageshack.us/img522/8110/ballanimationvraynewtg6.479bda0f95.jpg (http://g.imageshack.us/g.php?h=522&i=ballanimationvraynewtg6.jpg)

This one below with qmc as a primary bounce - longer and a bit differs.
http://img522.imageshack.us/img522/8765/ballanimationvrayqmcpribx2.jpg (http://imageshack.us)
http://img522.imageshack.us/img522/8765/ballanimationvrayqmcpribx2.ec549e631c.jpg (http://g.imageshack.us/g.php?h=522&i=ballanimationvrayqmcpribx2.jpg)

The system is core duo quad 2.4 and 2 gb of ram(1 machine).
I could tweak it further. There's also a problem with vraylightmaterial:it needs higher settings if used as a primary light source, so I had to make some settings 4 times higher than usually. I hope somebody will find a way not to do this. So if to compare with usual lights the difference would be more striking(up to 3 times I think).

This is your maxwell result to compare

http://img141.imageshack.us/img141/1773/maxwellfz3.jpg (http://imageshack.us)

I attached the file for vray users so they could tweak it better than me. I'm sure it's possible.

[sconlogue]

Hey Mister,

Nice results in VRAY! You are correct about the mats not being vary similar. When I have time a will rework the Mental Ray version and use the as close a shader as possible to the Maxwell mat. Also I think I left the "light ball" shiny plastic in Maxwell rather than stainless like in the mental ray version.

I rendered both with 8 dual core render nodes. The Maxwell version took about 12 hours, Metal Ray was 55 minutes.

I agree the shadows are too dark in the Mental Ray version, bumping up the FG bounces may help. I want to avoid actually reducing the shadow density but that may be requred as well.

[a23]

I agree the shadows are too dark in the Mental Ray version, bumping up the FG bounces may help. I want to avoid actually reducing the shadow density but that may be requred as well.

Are you using Gamma Correction (e.g. 2.2 Display Output) on the MentalRay rendering? Because that's what it looks like is not yet right.

Cheers,

Andy

[mister3d]

Thank you for this test, Samuel. Still I want to know: is your rendertime(6 mins for 1 maxwell frame and 26 secs for 1 mental ray frame) for 1 machine in your farm or all the machines rendering 1 frame? Because if it's at 1 machine I'm realy confused: we have to plus 75% of my time then:quad 2.4 is 75% times faster than one 3.1 duo(I may be wrong though: two 3,1 is 12,4, and 1 quad 2,4 is 9,6). I'm not sure if 2 additional gigs of ram make the rendering much faster.
But if we want really to compare the quality/speed, in maxwell there's still some noise at 100%, whereas it should be more like in my vray render.

[sconlogue]

Right, that was with multi-bucket on. So 23 seconds with 6 nodes rendering via MB.

[sconlogue]

Here is a revised Mental Ray render. This time I tried to match things up a bit better.

http://www.isclient.com/posts/MRBalls.jpg

[mister3d]

Very well! Could you provide this revised scene? I just wanted to make it in mental ray too, but you passed ahead of me. :) It would be interesting to look at your setup.

[sconlogue]

Very well! Could you provide this revised scene? I just wanted to make it in mental ray too, but you passed ahead of me. :) It would be interesting to look at your setup.

I just updated the Zip of the source files and updated the quicktime with the new setup. One thing that I messed up was forgetting to link the Area Light Plane to the box, so it doesn't match the Maxwell one after frame 0. Also the FG sampling isn't high enough to avoid some flicker.

[KV99]

I think that this 'comparison' stumbles before it begins, it is really, really difficult to compare rendering systems without making sure materials and lighting are set same for both systems. Currently, trying to render it myself I ended up realizing that it just won't work when I'm trying to achieve what this or that render already displays - especially if this or that render may not be as accurate as it should be.

It would really help to know what are the Maxwell material settings, light settings and camera settings. The original scene file doesn't have a single material even close to what's needed. Without having Maxwell, it's rather difficult to get to the point I'd like to with mental ray.

[mister3d]

I think that this 'comparison' stumbles before it begins, it is really, really difficult to compare rendering systems without making sure materials and lighting are set same for both systems. Currently, trying to render it myself I ended up realizing that it just won't work when I'm trying to achieve what this or that render already displays - especially if this or that render may not be as accurate as it should be.

It would really help to know what are the Maxwell material settings, light settings and camera settings. The original scene file doesn't have a single material even close to what's needed. Without having Maxwell, it's rather difficult to get to the point I'd like to with mental ray.

Most probably the only way is to visually compare materials. If materials would be the same for all the renderers, it would be easy to create universal libraries. But the controls are different.
My goal was to visually achieve similar to maxwell results, because they claim maxwell is unbiased megarealistic renderer. But what is interesting is that you get realistic results without much tweaking(I presume so), whereas in the other renderers you need more knowledge.
AFAIK, in maxwell you have a huge material library and you can't even tweak any settings.

[KV99]

Then you are not really comparing renderers anymore if basic properties of materials are different. It's not about controllers, physics of the material are still the same (as long as it's physics based renderer). But take the walls and the floor here for example, it has basic material applied to it. Maxwell treats it differently since it is- basic-.It applies physical properties to it that mental ray convert of standard max material does not, already paving a way to the situation where you'd try to achieve maxwell render by looking at maxwell render and tweaking as accordingly, but possibly going out of bounds of it's physical properties.

Since I don't have Maxwell myself and have never tried it, you're really saying that Maxwell gives no- control over what material settings there are? It just gives you bunch of materials to use and no material specific settings?

[mister3d]

What do you mean by "basic properties"?
You point is thus to compare the "clean" result without interfiering into the artistic side. But how will you measure the same conditions?
Gamma-corrected space 2.2
Same RGB values of all materials
the same reflectivity(but we must know exactly what reflectivity maxwell shaders use, and fresnel)
IOR
the same shader structure
the same specular level, anisotropy(and this is must be set by your eye too)
But how will you measure light intensity of light sources and light materials? I'm sure they are different, and here you must use your eye
exposure. DO all these renderers presume no exposure in these lighting conditions?
the same level of noise

Anyway, you don't use these mathematical measurements when you're making scenes. We are not mathematicians. We know there are differences(in gi mainly), so what's the point?
It's a very vague comparison, but the main for me is that you can achieve quite similar results using different renderers.
From another point, we could really know if there's a "magic" in maxwell with such a test.

[sconlogue]

Just to avoid any further argument, the actual point of this "Comparison" was not to 100% mimic Maxwell with Mental Ray but rather to see the performance differences for a REASONABLY similar result. Maxwell uses physical terminology and methodology in it's shading system so it is not even possible to technically and truly recreate a maxwell material in Mental Ray or Vray. Perfect reproduction was beside the point anyway. I am paying more attention to the overall look or feel. The subtle light casting effects of Maxwell is most interesting. I suppose they are not "effects" at all but rather light behavior we see everyday. In Mental Ray this is much harder to achieve but not impossible. Again we aren't talking perfect reproduction but rather reasonable results.

Render speed for certain lighting setups was where this all started. I have a number of projects that require "Natural" light phenomena and animation as well. Maxwell gets great results but the render times are high. So I thought about creating a reasonable comparison with Mental Ray and seeing if I could achieve similar results with improved performance per frame. Certainly they will never look the same. But if I can get close and save hours and hours of render time, great!

So let's not get too stuck on this.

[sconlogue]

Oh ya, For those of you without Maxwell. You can always download the demo --> http://www.maxwellrender.com/

And regarding light setup. You will find that since Maxwell only does "object lights", meaning you apply a self emitting mat to geometry and now you have a light, light position and size have a slightly different meaning. That's why I went with testing the scene with area lights in Mentalray, since spots are based on points of light and would be completely different. Area lights are close enough. Though I know in the new Mentalray shipped with Max 2008 you can have geometry based lighting. I have to give that a try.

[mister3d]

You could have a self-emitting surface in Mental ray in previous versions
http://img512.imageshack.us/my.php?image=glowandglareym6.jpg

Create mental ray type material(instead of standard max) and put in surface slot- glow(lume). Then adjust color and intensity. The light from glow is seen with fg and it affects shadows, i.e. objects generate shadows from it. So it's what you need. And for neon lights too.
If you want to add specular to it, then put in glow surface material either dgs, or archmat if you use 9 max. If you use dgs and want blurred-flurred reflection then set glossy value it will be rate of reflection(you can put fresnel falloff there) and make your shiny value like between 10-1, these are values when reflection begins to blur.
You can also use parti-volume shader in volume slot of mental ray material and put in surface slot transmat(which makes furface invisible but you can see volume light for this material if applied, it says it transmits information further and won't show any surface information). On Jeff Patton website there's an explanation how to use parti-volume shader, though there it used in volume light example, but you can get the controls. As fas as I remember scatter color is the color of light and extinction controls the brightness(may be wrong), so put it like o,oooo1 and it will be bright. If it's minus it's black, and if make values bigger will become darker too... So I would stick to glow.

[sconlogue]

Hey Mister,

Thanks for the tip. I have been using a slightly different method, I am not sure the results are exactly the same though. I have been taking a MR mat or Archi/Design mat and applying "output" to the self ilum. Then add bitmap or gradient or anythings to get the color. The just adust the level of the output and you can control how much light it emitted into the FG scene.

In the new MR I think you just have the option in ArchiDesign to make it an emitter. You can even choose the light type and number of numens etc... just like in Maxwell! Sweet!

[KV99]

Yes, there are methods to fake illumination glow in previous Max releases with mental ray. One of them was the abovementioned output and glow shaders in the illumination slots, the other was unhiding the shaders that allowed objects to be light sources.

Max 2008 does have an actual Glow options in A&D material, though I'm not overly excited about it. I've had some strange renders with it and using it as a main light source can bring more unsatisfying results than regular lights. On the plus side however, it does have options for physical light colors and physical intensity (in cd/m2). I guess I just need more time with it to get full functionality out of it.

[KV99]

Well, as I said before, I'm not really keen on considering any sort of 'versus' between rendering systems, but I gave it a try to mimic something in between the renders displayed before in this thread for vray and maxwell. Im not too happy about the lighting nor the materials of the balls themselves, but I am quite glad about beating both rendering systems in rendering times (and that with depth of field enabled - something I considered a must to follow maxwell)

http://www.waher.net/mrtry.jpg

I set up raytracing based on what Maxwell did, for some reason the Maxwell render has only one raytrace depth in use, which is the reason why balls in the reflections don't reflect the glowing ball.

[mister3d]

KristoVaher, what a nice quality and speed! Very precise GI! What optimisations did you use, if you want to tell us? :)

[KV99]

Of course, I don't mind explaining what achieved this result, I would not just post it here for show :) It's really alot of trial and error here. First thing was to make sure that I used tonemapping and gamma, I used that to even out the general look of the image what maxwell displayed (this minimally washed out colors look). With tonemapping I took as much out of the shadows as possible, since I didn't know the exact parameters of the surrounding environment (the white box) then I really didn't have any other choice but to try and achieve the similar look through faking it.

This means that I ended up tweaking the balance of the balls light and the plane's light to mimic what Maxwell displayed. I went out physical and default settings to achieve that. What I used for cue was the shadows one of the ball casts on the wall, and the bottom-most ball and the ground. I also desaturated the colors on the image slightly, and moved the whitepoint of the tonemap a bit towards achieving this slightly yellowish color instead of white.

I did not use any Global Illumination nor caustics, neither would have made sense since the environment relied strongest on the glowing ball and area lights. One of the biggest problems was getting the shadows display correctly, since the material seemed to be very sensitive to shadows in mental ray, I needed to fade the shadows more. I did that by adjusting the shadow color of the area light and also added very very minimal glow to the box itself (sort of simulating a very small bit of extra global illumination which I did not use).

Another thing that's important is minimal depth of field. Im assuming Maxwell always renders depth of field, and every camera shot naturally has it to some extent. I used it to slightly smooth the background and the ball that's farthest away.

Since Maxwell did not have a deep raytracing, then - as I said before - I limited raytraces to slightly boost up rendering times.

What I wasn't happy with was the lighting of the back wall, which in Maxwell is almost not visible, but those little details didn't bother me that much. I'm not happy with the settings since they are not physically correct, but some of them are here:

Tonemapping and Exposure:
Exposure Value 2.0 (very, very low, it's so as a result of light settings)
Hilights and burn 0.18
Midtones 1.010
Shadows 0.1 (taking a bit out of the shadows)
Saturation 0.8
Whitepoint ~7000-8000 Kelvin
And some slight vignetting (part of Max 2008 photographic exposure)

Material (I was least satisfied with the ball material itself)
Reflections 0.75, glossy 0.8
Diffuse level 1.0 with brownish color
BRDF 0 degrees 0.12, 90 degrees 1.0, Curve shape 3.46

Glowing Ball light material
Kelvin 6500, 20000,0cd/m2
Does not affect FG! Was only put in place to represent the glowing circle of the ball

Area Light
Color D65White, Intensity 51cd (!!! as I said, physically rather strange settings, I kept these settings small since they kept overtaking the glowing balls light)
(I also used an actual white illuminated plane here that didn't affect FG, for the reflections to be more accurate)
Also, light affects shadow color was set, as well as shadow density reduced to 0.8

Glowing ball light
A physically incorrect mental ray spherical area light added -inside- the glowing ball. mental ray doesn't have true object lights, so I had to fake the glow's light effect. I made sure the white glowing circle didn't cast shadows and the spherical light inside the sphere was what illuminated the scene. It's parameters:
Intensity 17,122 (!!! again, not physically correct at all, not even close)
Inverse square decay that starts outside the exterior of the glowing ball sphere (mimicing that the light doesn't come from the center but instead from the circular illuminated part of the ball)

White environment box material
It's a white material with roughness 1.0 and glows 0,159cd/m2. It's a very minimal effect that I deemed necessary since the shadows were a bit too dark for me otherwise, and it somewhat faked the global illumination I was missing.

Oh and final gathers settings were quite low, tweaked a bit from the "low" template. Reason for this was quite simple, the scene has very little detailed illumination and shading, it benefitted from simply using lower settings than higher ones speed wise.

Hope that helps. Remember, since it was literally impossible to get physically correct renders here in such a comparison, this solution relies entirely on the artists eye and depends on one situation alone.

[mister3d]

KristoVaher, your setup is somehow similar to mine, except that I used vray lighmaterial, but it would be better to use an area light inside the ball as you did (it would make a more precise GI and would render faster). Your understanding of Mental is impressive.
But what this test showed to me is that neither Vray, nor Mental ray can't reach a realistic result "out of the box", whereas Maxwell IMHO can. I believe(based on what I read here from archives in this thread) that Maxwell has a more accurate material simulation, and this is the reason.

[KV99]

Of course mental ray won't reach realistic results out of the box. It's a production renderer system, it's not meant to tie users up. mental ray allows you to get whatever result you want, from realistic to truly artistic, that's what's its strength. From my personal experience, it's the most versatile rendering system out there.

And it is alot easier to reach a realistic result 'out of the box' if all of the properties in the scene are same or similar to that of Maxwell's. But in this case, it seemingly was not possible and to get the same result, it needs alot of trial and error. It would have been easier if I knew alot about Maxwell and thus was alot more capable of converting settings from a physics based rendering system to mental ray which simulates physics.

And in my opinion it's very good that you can't reach realistic result out of the box just like that. It teaches the user alot more about how to achieve a specific result and in the end - how to achieve the result the user wants. Maxwell is really good to be used as a reference when trying to achieve something with mental ray that's close to photoreal, but I have yet to purchase Maxwell just for that reason since the pricetag is rather hefty for a really restrictive rendering system I'd use mostly for reference.

[sconlogue]

KristoVaher Nice work! I haven't tried the scene in 2008 yet. I'll have to start it up and see what happens. Though I think the result you have here is the best we could expect to reach for the reasons you both just mentioned regarding Maxwell's physically accurate methodology.

Is there any chance you could render out the full animation with these settings? This is the other point of the comparison. So get Mental close but also with the condition that the animation looks as natural reasonably close. Since you didn't use GI or FG here flicker shouldn't be a problem. If you don't have the ability to render it all out you can post the .max file and I render it on our farm.

[CGTalk Moderation]

This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.