Why v-ray is better than mental ray

[ramy]

Ok. I've been working on an animation for the last several months using mental ray. I've been really happy with the lighting results, but totally frustrated with the sampling. I switched over to unified sampling in mental ray hoping it would work faster...it did, but even with the noise in the image I was still getting 30min / frame @ 800px.

 

It was just too slow for an animation. So I thought I'd switch everything over to v-ray. WOW. Off the bat I was getting crazy fast rendering times with very comparable results.

 

I will say that the materials and lighting in mr are more "physically accurate", the results in v-ray are great considering the render times.

 

Here are my results...

 

 

If you want to know more about my sampling tests here are the links:

DMC vs. Unified

Unified sampling

 

happy rendering!

[rafaellopez]

Hello Ramy

 

Thanks for sharing these point or differences between Mental Ray and Vray. I started with Mental Ray, has the same issues too. I now work with Vray, but I am considering to work back with Mental Ray. Thanks for sharing. Love your work.

[jimbuchanan1]

Hey Ramy,

interesting comparison - thanks.

Have you tried iray? Easier to set up than both, but needs some serious graphics

card(s).

Check out Jeff Patton's site: http://jeffpatton.net/

[Chris MacDonald]

I too started with Mental Ray, and I didn't switch because of the render times, or even image quality really because I think it's just as capable. I switched because when I hit my memory limit in mental ray (which was 6gb at the time) then it would crash the render. Vray however just carries on (albeit a bit slower because it has to cache things to HD, presumably) and means that I never have to worry about coming in to work in the morning to a failed render. THAT is why I love VRay... Oh and that people can suggest something on their forums and they go away and implement it ASAP! Wonderful tech support.

[petesedlacek]

I'm new on this forum, but what in the world is causing that MR render to take 30 minutes? There is no way that should take more than 5-10 tops, with no noise at all. Are you trying to brute force the Final Gather? I would love to get my hands on the scene and check it out.

[Justin Hunt]

I'd also like to see your mr set-up because other than the unified sampling issue there is something funky going on. I suspect its the lights, especially the portal light?? and the pendant lights above the counter.

 

I know Ramy does know how to set up a mr scene efficiently so its more out of curiosity than anything else.

 

jhv

[ramy]

Guys thanks for the comments. I did some test scenes for testing sampling between mr unified sampling and vr DMC and in a simple test scene didn't really notice a difference. However in a production environment (the example above) they were no where close. A little about the renderings on this thread:

 

MR image: GI and FG were precalculated, so render above is reading them from file. There is 1 mrSkyPortal (above opening on left), and the rest are photometric lights; The rectangle lights above each have 2 shadow samples. The pendant lights are actually point shapes. 3 ies over the menus. Sampling is using unified with min:1, max:100, quality: 0.5, threshold: 0.04.

 

VR image: LC and IR were precaluclated, and render is reading from file as well. All lights are vray lights. 1 rectangle light for opening with outdoor setting. Rectangular lights on ceiling are straight forward. The pendant lights are actually vray light spheres. Sampling is using Adaptive DMC with min:1, max:6. DMC sampler at default values.

 

If anything, based on the lights, the mr should go faster. This is a huge scene with a building that is modeled inside and out, with 3 floors and alot of entourage. The only thing that I can think of is that vray simply handles what is immediately at hand better than mr with really large scenes. Was really surprised to see this as I've been using mr for years and usually get comparable results. Guess not this time.

[Justin Hunt]

which is why I think there is something funky going on.

[Nic H]

no real surprise.

 

mental ray is autodesk so doesn't get developed at all - if anything it gets slower and less flexible.

 

vray is actively developed by a responsive and talented development team resulting in a very flexible and fast product.

 

that said i couldn't tell the difference between them visually in the right hands and im not sure a small render speed difference on one scene should sway your choice between the two.

[thorsten hartmann]

Hi

 

the answer of your question is in your images. You have knowledge about Vray, and it looks good, but have not enough knowledge about mental ray and it looks bad. Your Setting about Unified Sampling is wrong. look here at my easy Workflow: http://www.infinity-vision.de/blog/unified-sampling-at-3dsmax2013

 

mfg

hot chip

[ramy]

Thorsten, even with your suggested settings, I'm still getting very slow renderings with unified. Even with a material override of a plain grey A&D material the render times were just too slow. I'm not using DOF or moBlur, so it seems unified is not the best solution for this situation.

 

I will say from my tests that when you do include DOF and moBlur in the render unified is much faster than anything else.

[thorsten hartmann]

Hi Ramy,

 

we had do many test vray vs mr, and we think vray is not faster, or mental is better. The knowledge of artist is very important.

[williamrav]

Hello Ramy,

 

I have to go with Thorsten here, there's something very fishy with that MR scene, the amount of noise you are getting is not normal at all, I too suspect the lights plus their sampling, I also see some overexposed areas plus problems with the overall ligthing of the scene which could use more work if you don't mind me say, and not enough bounces in some places (such as the upper wall section of the green lights, it'll be interesting if you figure out the issue and let us know about it.

 

meanwhile regarding the comparison:

while I have been using both renderers for a while now, I must say that MR is not slower than vray or vice versa, in fact after testing production scenes with vray we had to switch back to mental ray to get them out the door, while vray probably works best for architecture viz scenes with baked lights, its hardly cutting it for us when it comes to real production scenes such as displacements plus 3d motion blur and characters, I have no idea what the big fuss was about with vray having faster 3d mblur than MR ( this was true before unified sampling was introduced to MR), we couldnt render flicker free aniimation in vray on characters and small details without having to switch to brute force which practically killed the render plus gave us lots of noise, havin Mblur on top it just went on for hours on a simple scene with displacements and one area light!

 

having to use baked lighting or interpolation is out of question for that still causes flickering and loss of detail.

 

MR has its own issues with final gather but there are proper work arounds which mostly solve the flicker issue plus with unified sampling 3d mblur i can safely say is faster than vray from the tests we've done here.

 

as you also mentioned MR's shaders seems above all else, for us the new skin shader in max 2013 alone is worth the consideration.

 

one other big advantage of Mr is the ability to "cheat" and be more flexible with your scene and shaders/lights, vray is all too physical and that takes more time to render even on simple scenes, so you cant go around it that much.

 

MR still has issues just as any other renderer out there, and as many mentioned here its up to the user not the software.

 

I think Mr is picking up though, slowly but it would be gettin there soon, i think mental ray 4 will be a great release with new GI engine, and all the other gadgets built in.

 

regards,

[issacdhan]

Ramy, I would love to render your scene.

Edited September 6, 2012 by issacdhan

[williamrav]

by the way Ramy, its a good idea to put up these test comparisons with unified sampling, i think it could be helpful to many.

 

regarding motion blur, unified works quite well, unfortunately depth of field has always been problematically slow in most render engines out there, therefore personally we always avoid it in actual engines and post it later, the thing which would make it more accurate now in MR is that in post the z-depth channel is now exported with proper 3d mblur which before unified, was very difficult to achieve correctly and in a relatively fast clean render times since 3d mblur with mental ray was a real pain, the 3d mblur would be applied on the zdepth output itself and make the post brokeh DOF that much cleaner.

 

there are a few things missing in MR/max camera regarding DOF compared to vray's physical camera, and that is:

 

when you rack focus there's a slight change in perspective deformation that happens in real lenses/cams which is also present in vray's phys cam, that seems to be missing in max's camera, some sort of a lens script or camera script should fix this, if you guys know of any please point through.

 

thanks,

[kazemghari]

Hi remy , i too render my scenes with both mental ray and Vray . i actually agree more with william. i usally render my still images using Vray cause i with less of a work i get better results though i render my animations using Mental cause i found it faster.

here is 3 tip that might speed up your renders using Mentalray:

1.always check the far attenuation box in photometric lights, if you haven`t done yet i imagine you would be amazed by the time change in renders.

2.in animations try not to use sky portals and simulate the effect someway else, sky portals are really great but they do all the calculation by themselves and you really don`t have much control over them, but if you have to use them use them with mib-blackbody shader, that would turn the calculations faster

3.i believe the biggest diffrence between mentalray and Vray lies in materials, for example the refraction in mentalray is really a time killer, master materials in Mentalray and it would boost your render times

[thorsten hartmann]

ok i test the scene and it give many to do....

 

- all peoples of the World. We use unit-system "METER" and not inch, zoll or what ever

- put the sun in the middle of you scene

- use photons and set sun photon to 30 meters radius

- render settings autovolume on

- place holder objects "on" only if you have not enough ram

- windows glasses objects properties = no cast shadow, fg invisible, if frames and glas single objects must you detach glas elements to new a other object.

- if you use photon, must you enable gi photons of the lights

- photon emitted amount render setings 100.000 Photon

- for each Licht-Group of instances use 100.000 Photon (Render Settinges), go to Light-properties, engergy-photon and set th GI-Photon to 1/light_isnstances_amount

 

i have only spend 1/2 hours to optimize.

 

 

@rama: you become a e-mail with include the scene.

 

mfg

hot chip

Edited September 2, 2012 by thorsten hartmann

[ramy]

Thanks for all the tips and input everyone. A couple of things about rendering speeds. There are only a few things that actually control affect the efficiency of rendering, so here is a list of what does or does not affect the rendering, in mr (vr does have light maps that affect the rasterizing process such as glossy reflections in light-cache, and light shadows in irr-map):

 

-Any pre-calculation of Indirect Illumination (ie: GI/FG), do not affect the rendering speed (the rasterizing process), assuming that this is pre-calculated and read from a cached file. It makes no difference if 50 or 500 Rays per FG point were calculated. This also applies to the Energy Photon settings in lights.

 

-Things that do affect rendering speed: materials (textures, reflections, refractions), shadow samples (point shadow vs. rectangle vs. ies), of course renderer sampling settings / filters / spacial contrast, and how many cores/buckets you can get

 

I found the major killer in my mr renderings as someone pointed out are the mrSkyPortals. Even with the mr_blackbody (this simply controls the kelvin color of the light), they are great lights with great shadows but big time killers.

 

I managed to get this 800 px image to render in mr at 9min15sec with unified sampling with a couple of tricks:

 

 

A work around that is somewhat of a trick is to pre-calc GI/FG with mrSkyPortals on, then turn them off when it actually comes down to rendering. Sure you don't get the direct illumination and shadows, but you still get the GI/FG color of the sky put into your scenes.

 

I also applied this trick to the rectangular phtotometric lights. When it came time to render, I turned them off and unhid a replacement geometry with an A&D glow applied. This cut down on the shadow samples completely and made the render much faster, still using the GI/FG from the original photometrics...very unconventional.

 

Overall, if I don't have moBlur or DOF in the render, I'll stay away from unified and stick to standard sampling.

Edited September 3, 2012 by ramy

[thorsten hartmann]

hi ramy,

 

mental ray can store the photons from portallight in a photonmap for fg-calculation, but you need the original skyportallight shader. Now the arealight sample from portalight is off for beauty rendering and mental ray use shadows only from the photons. Photons are physically correct and work with Caustics too. You have the best result. Vray have photons, but in vray you store the lightsample in irradiance map.

 

Edited September 3, 2012 by thorsten hartmann

[ramy]

interesting...i've never seen this before...i'll have to take a closer look.

[ramy]

All,

I finished the animation to this project and kept it in vray. Thought I would share: http://www.ramyhanna.com/2012/10/wul.html

[Justin Hunt]

Came out nicely well done

[thorsten hartmann]

Hi Ramy,

 

can you post it on a other place, for example vimeo.com. I can´t read it in germany, it is permission denied.

 

mfg

hot chip

[dhammikaherath]

Nice stuff Ramy..

[ramy]

Thanks. Thorsten, just updated it to vimeo.