Mental Ray: worth learning?

[Nizar]

Hi Rez007,
as I wrote, thanks for your contribution and your time.

So far it seems that you like when people give you advance and where to find better information on how to learn Mental Ray, but to me, it feels like you keep holding on/agreeing to the statements of Mental Ray being "bad", basically to reaffirm your own feelings about.

The idea was this, try to understood where I did wrong and why people like or using mental ray. Why is so strange if agree with some statement and share a negative felling with others? Facing mental ray is intimidating and time consuming, but I stress my previous sentence with the note "I have not the experience", so I cannot fully judge it. I retired after a "big" tentative (first time I struggle with mental ray for three months, without arriving to a satisfaction result for my need), now I have a entire month free for testing it and want try it again, the reason, as I said, was because I have just purchased it, want doing all inside softimage and want understood how and why many expert users love it (and how obtain this great result). Cycles and modo pleased me, I think they are good, but sometime I fell like need a bit more control, in this way I will understood if I really need it or not.

From the above statements, seems impossible learn softimage in a month, but I'm not totally new to it, so want try again. Despite the new rendering actual disposable, mental ray continue to be used in many big production, must be a reason. Whit this don't mean is a no brain choice (is used for many factors, not less important the presence of many great artist used to it)

Thanks Maximun, I will read with interest the link you suggested and also will read Elemental Ray blog.

[Rez007]

Hi Nizar,

I think if you look over the material on Elemental Ray and also some posts people have written here, you might be quite surprised that is actually not too hard to get up and running on Mental Ray. I think you might find it pretty nice as well, especially if you are looking for something that has more flexibility. Sure it has some limitations, just like all renderers do, but I think if you come into it with an open-mind and learn Unified Sampling and how it works, you should be good to go. I guess at that point you can figure out if it is something you want to invest more time in or not. If you have any questions on the way, I am sure myself and others would be happy to help you. Best of luck.

[remydrh]

The point of my last explanation is that similar techniques are used in other renderers, including Vray.

The exposure of these controls is different.

For instance, DMC versus Unified Sampling. Or Irradiance Mapping versus Final Gather. Dome light versus IBL. Detail shadowmaps versus deep shadows. GI versus photons. Etc.

What is usually lost in these conversations that I am trying to point out, is that the names may be different but the underlying strategy of these methods are the same. So understanding Final Gather relates to the Irradiance Map. Understanding Unified Sampling relates to DMC. Rasterizer relates to Reyes. The claim that there is no equivalent in mental ray is not factual. This is why I both have the blog and comment on this post. Not because I am trolling but because I am correcting the information provided so that you may make a more balanced decision. (This post feeds to the blog so I feel my presence here isn't uncalled for.)

I have not made judgement calls about which features are "superior", "suck", or are "worthless". Instead I have provided the comparison for you to see between the features and their nomenclature.

These are all the same techniques using different controls. (As a sidenote: brute force has been around since at least 1996 (Veach). This means brute force methods are actually less modern, not more. But since CPUs have become so much more powerful, the need to use interpolated techniques is less obvious. If 6 hours a frame was acceptable last year, then 6 hours on a newer machine this year is still usually acceptable even if using brute force. Typically referred to as "Blinn's Law")

So what's truly different if you choose to learn another renderer? Well, integration is different. This is where mental ray complexity can become a bother. The documentation isn't particularly great from Autodesk. And the documentation from nVidia is designed for integrators. This may leave you to the mercy of the internet.

As for performance, at least time per frame, if you go through the steps of baking, Renderman probably still outperforms other renderers for actual render times. And since experience for time per frame varies greatly (recently I rendered a shot with 40 area lights and motion blur in mental ray at 1 hour a frame) I would say that's up to the user's usual work subject to decide. Lots of hair, or area lights, or motion blur, etc. This will affect what renderer has the control you like more.

Sidenote on Chad's blog: What he fails to mention (and I was there when this all occurred at Luma) is partially covered under an NDA (although he does mention a few things that are still covered albeit in a sideways sort of way). It's a little like asking the wolf about the missing chickens. But I will say this much: the pipeline at Luma was using deprecated code that was not updated. In fact, some bugs in the code being used had been reported in their framebuffer system that causes double render times and up to 3X the memory usage when outputting passes. This was not a failing in the mental ray code, it was the Nexus Framebuffer passes using the old mechanism that was updated in mental ray 3.6+ Unfortunately there was no desire to update the code or maintain the pipeline despite its existence for more than a year. The method was the same that Autodesk had unwittingly used in their first attempts at using framebuffers in Maya. You'll have seen those comments where people reported exorbitant render times in earlier versions of Maya. I believe XSI gratefully escaped this fate for some time.

And the ultimate recommendation: if you want to learn something, find someone that will teach it to you that has the experience to do so. If after that you do not like the solution, then don't use it.

(Note to moderator: I have provided factual information that can all be cited or otherwise gleaned from appropriate sources (papers and/or user documentation) without the use of expletives or judgment calls as noted above. Thanks for keeping a watchful eye and I will minimize my sarcasm in the future. )

[Kzin]

maximus:

you have alot of options to do brute force in mr.
the easiest (and slowest) would be exact fg.
then you have brute force ip's.
in addition, if one bounce is enough you can use ibl and ao colorbleed.
and last but not least, path tracing with megatk shader (there is a path trace mat, but it has poor control). with it you have a arnold like shading and lighting if you stay with physical lights only (without the bsdf parts).

what you say about unified and that you need to get thru alot of tests to understand how to setup it correctly: i had the same problems in vray with dmc. you have to run alot of tests to get a good quality without render ages because you oversample everything. have you tryed vray brute force with alot of glossis in an interior? rendertimes are ridiculous high to get smooth results. because of this, you read everythere that people using this mode still have noise, but its ok because it looks like filmnoise (the same thing for arnold). and i also heard from different well known places which using vray that they using ridiculous fixed sampling settings for all project without any tuning. simply because they have a big renderfarm and they dont care about rendertime. throw in the frames and render 20 hours per frame, no problem because we have 1000 rendernodes. decent gi solution is a different thing from my point of view.

the biggest mr problem in xsi is the integration from ad and the bugs resulting from it. strange behaviour of ad mat (always use mr's native mia mat), the geo shader thing, not working standins make it really hard to work with mr. in addition, bad default presets makes it hard for newer users. then you have users which are not willing to read the docs and complaining about things which are in mr but not as default visible like the whole lighting shaders. all wants the physical approach with lighting but they also want to use a spotlight, mmh. ;)

and i will now write some words on the "no tweaking needed for arnold" phrase you can read everywhere. from my own experience it only half of the story. the problem with arnold is the really high rendertime because of its brute force nature. we have no big renderfarm here to compensate the higher rendertime so we have to tune the scenes to make it render faster. that means you use all the tricks you know from mr or vray to make the rendering faster. the problem in arnold is that some tricks are not that easy because arnold is not designed to do these tricks.
from my experience, brute force in general is to slow (on cpu) at the moment for everydays usage except you have a big renderfarm behind you.

[Kzin]

In mental ray I need to set a property value into Scalar Change Range (at least, need only to edit the two parameters New Range start and new range end, is more simple when you know the import value from zbrush), but not only I need to cage, with fine setting, the Geometry Approximation values (Polygon mesh subtab (angle and sunbdivision (ogl and render level) and Displacement (where can select Fine, parametric and Lengh/distance/angle), in "Fine" I have five slider to setting and one of the most important, lenght, can arrive to absurd value of 0.002...).

they should copy and paste the explanation from the "rendering with mental ray" book for things like this. this shows one more time that people dont understand WHY these settings are in. i highly recommend this book because it explains alot of things really good and i wish mi would make a doc file out of it and put it together with the mr doc for free (will not happen because of the copyrights i think).

[Kerro Perro]

and i will now write some words on the "no tweaking needed for arnold" phrase you can read everywhere. from my own experience it only half of the story. the problem with arnold is the really high rendertime because of its brute force nature. we have no big renderfarm here to compensate the higher rendertime so we have to tune the scenes to make it render faster. that means you use all the tricks you know from mr or vray to make the rendering faster. the problem in arnold is that some tricks are not that easy because arnold is not designed to do these tricks.
from my experience, brute force in general is to slow (on cpu) at the moment for everydays usage except you have a big renderfarm behind you.

I have to take exception to the phrase "really high rendertime" for Arnold. I've worked with it and yes it does still require some tweaking - i won't go into the how's but that will be public soon enough - but it's still faster than Maxwell and Vray with similair scenes. Also the studio i worked at last year had it and they just set up scenes on regular workstations and rendered them out overnight on a 10-node renderfarm - wich i wouldn't classify as big.

True if you really know what you're doing and you set your scene up perfect you can be faster in Mental Ray (i'm a fan of it remember ;) ) but your making it sound like some unusable overblown renderer and it isn't.

Btw there was a great thread on here last year where a guy compared a big scenes in MR, Vray and Arnold - with AT-AT models - wich may be worth looking at, but i'm too lazy to go searching for it

[Kzin]

True if you really know what you're doing and you set your scene up perfect you can be faster in Mental Ray (i'm a fan of it remember ;) ) but your making it sound like some unusable overblown renderer and it isn't.

i dont say that, but its hard to discuss about it when people which dont know arnold writing about it like its a miracle thing which do all the work for you.
it does some things better: streamlined xsi integration with all the modern features only (not the old school default shaders like in mr), its "activate it" and "forget about it" gi.
the funny thing is, in mr people dont want to learn the tech, but in arnold you also need it because you have to know what you do.
the grass is always greener on the other side, learned it by myself. but if you are in neighbors garden, you will see that your own one is not less green. ;)

[Kerro Perro]

True if you really know what you're doing and you set your scene up perfect you can be faster in Mental Ray (i'm a fan of it remember ;) ) but your making it sound like some unusable overblown renderer and it isn't.

i dont say that, but its hard to discuss about it when people which dont know arnold writing about it like its a miracle thing which do all the work for you.
it does some things better: streamlined xsi integration with all the modern features only (not the old school default shaders like in mr), its "activate it" and "forget about it" gi.
the funny thing is, in mr people dont want to learn the tech, but in arnold you also need it because you have to know what you do.

Totally, you really need someone to walk you through it because it has some unique stuff you need to pay in mind.

the grass is always greener on the other side, learned it by myself. but if you are in neighbors garden, you will see that your own one is not less green. ;)

Yup totally agree with that as well, as i said in an earlier post i wasted a lot of time obsessing over wich renderer does what - and now i (usually) use techniques in mr that are over 10 years old and they work perfect for me! ;)

[Maximus]

One thing might not be clear, I love mental ray and i still use it. What i completely refuse to accept is the total disaster both Autodesk and Mental Images did with this product.
I dont care if they throw the ball each other, i'm an user that found a product completely fucked up from the ground up.

I hear what you say both Kzin and Remydrh and i agree mental isnt bad render, but both AD and MI made it counter productive to use, because of its integration in 3dcc packages, because of slow updates in 3dcc packages, because of poor and confusing documentation, because of poorly explained techniques, because of legacy stuff let rotting for ages, because of a lot of other things i've already listed.

You listed couple of things for Mental ray and its bruteforce. (beware that i spent months to not say years on everything you already said so i'm not that new to this).

- Exact Fg- Render times makes this completely unusuable for any kind of work, even if you use it with Unified sampling and lowering the rays value of FG its still incredibly slow, might as well just start Maxwell render, and i'm not Joking so if i have to use this i'll just use Maxwell.

- Brute force ip's - Brute force IP are completely full of problems, artifacts, bad implementation in Softimage, last time i used it Softimage crashed in creating the IPmap because there was too many importons i guess. Also you can check this thread because its the same that happened to me (and is also the reason why nobody uses IP)
http://forum.mentalimages.com/showthrea ... -lightning this guy had a 9 hours rendering for a CLAY RENDER.
To me that just sounds utterly stupid render time. And still he had artifacts, shooting like 1 milion+ IP. Lets face it, its a tech that is not working. And even if it might (on super rare occasions) you have to tweak a gazillion of parameters (as usual).

- path tracing with megatk shader - never tried this but i tried the built in path tracing material which i should refrain to comment because its gonna just be offensive. I wonder who came up with a shader to calculate path tracing instead of a GI system but lets move on. I will give it a try i guess?

Now compare all those techniques you told me with 1) Vray in which you just activate brute force and set samples 2) arnold in which you just hit render and set samples.

Its not my fault, neither yours if we are in this status of stagnant implementation, everything is fucked up inside Softimage. You ever had a dense a complex scene with lots of materials? Try to open material manager even with apple preview set to draft most of the times it crashes because it starts rendering all the shaderball. Yes you can disable the preview render but i damn need the preview. Am i supposed to disable everything because MR/AD/Softimage are uncapable of optimizing things?
Maxwell doesnt start rendering material preview if you dont hit "Refresh", it leaves you the last render as preview. Those are small things but you dont need einstein to optimize and get things better.

Rendering is being overlooked since a lot of time now inside Softimage, the current status is really insulting for professionals that use this software. And things arent going to change.

[Kzin]

Now compare all those techniques you told me with 1) Vray in which you just activate brute force and set samples 2) arnold in which you just hit render and set samples.

have you ever used vray's brute force? its slow like hell and unsuable except for some outdoor shoots where noise is acceptable.
interior with brute force is not usable if you need clean solutions (especially with brute force glossis). pt with megatk shader is comparable to vrays brute force. pt in mr could be ok for outdoor, like vray's pt, but forget booth for indoor renderings (portals with brute force ip's are not that slow, but cant say something about animation stability at this point). the only renderer which is production ready with brute force for indoor is arnold. but this does not mean its fast. rendertimes of 10 hours and more are normal here so dont think its better in another renderer. only gpu renderer like octane CAN help, but it depends alot on the scene. you can render interiors in 4 hours, but also in 20 (take a look in the forum and dont be shocked by the rendertimes ;)
). from all my comparisions, portals with low fg settings in mr are not that bad, they perform very nice and they are animation stable. with aocolorbleed you also get nice indirect bleeding and shadowing details. but, we need a new gi engine here, i dont say all is perfect.

[Kzin]

Its not my fault, neither yours if we are in this status of stagnant implementation, everything is fucked up inside Softimage. You ever had a dense a complex scene with lots of materials? Try to open material manager even with apple preview set to draft most of the times it crashes because it starts rendering all the shaderball. Yes you can disable the preview render but i damn need the preview. Am i supposed to disable everything because MR/AD/Softimage are uncapable of optimizing things?
Maxwell doesnt start rendering material preview if you dont hit "Refresh", it leaves you the last render as preview. Those are small things but you dont need einstein to optimize and get things better.

these are different things which have nothing to do with mr. its the bad implementation of the features and i am with you here. i use the mat editor only rarely because of the bad performance (same for rendertree).
like the new hq viewport, its a bad joke and this shows that there is no qa at ad. nobody is using the features at ad. because of this i call the last years features, feature list features. they are good on the their list, but mostly unusable. ad looses the contact to the users. they started to change this for maya because of the big studios but i dont have high hopes for si here.

[Maximus]

can you please link me a working set of those megaShaders? compatible with si2013 x64 possibly, cause i didnt manage to find them. I wanna give it a try

[Falam]

Hmm that is interesting because the flicker problem is pretty well known so if you have a simple method that works a think a lot of people would be very interested

Yes it is a well-known issue, but there are others who have generated flicker-free animation using the same technique that I have. Maybe a lot of people keep just thinking if they throw in a HD Reflection map and use that to derive an FG output that all will work - that might be one of the main problems. All I did was this 1.) Unified Sampling (I don't know for sure if this has a great effect difference from Adaptive Sampling, I just tend to use this method and no more AS) and 2.) Use a blurred image to calculate FG. and 3.) I believe I have the FG algorithm set to "Multiframe" - I am not in front of Softimage right now, but I thnk that is what it was set to. You can take it for what it is worth, but I am just saying that this is how I get flicker-free animation...and then of course which is mentioned in the links provided, I guess you can brute force FG if you have the time...

Thank you for the tip about Unified sampling, I can't believe I over looked that option, I usually see it and try it out later, and I didn't even see it, sheesh. This thread relates to a similar thread I started about semi-realistic reflections using the Arch shader, because it deals with Mental Ray I'm rendering a still image using FG,GI what do you mean by a FG map, currently the render is slow, and I'd like to speed this up.

[Falam]

Thanks for so many info Rez007 I will try tomorrow and post here some results. I have an entire month for learn mental ray.

About Gustavo tutorial, I have the first one (demystify Render Tree). I think is a good tutorial and I like how he explain his workflow.

Falam@ modo render can be mastered in about a month (or lesser). Modo render output is very good IMHO. The book (a real bible) you mentioned is out of print, but is also outdated (correct me, but don't talk about unified sampling and architectural material)

I never used Modo. I'm sticking with Mental Ray, I've come this far, best to just master it which I'm not there yet

[Falam]

In my opinion you should not really worry about that. Mental ray with unified sampling is so easy to set up, actually I find it quite easier than Vray ( perharps not as effective sometimes ). From my point of view I just care about:

1-. Unified Sampling

2-. Mia_material_advanced

3-. Spot, direction and portal lights

4-. Most of the time I use FG so depending on the project I use HDRI.

Guess what, I do the same in Vray. What I find easier in Vray? that all the passes and everything within Vray work perfectly together and there are presets for everything. For example, yo can just set the passes with Vray much easier than with MR. That's my point of view. Other advantage of Vray, by default is faster and with the proxies you con optimize renders quite a bit.

I'm just saying that you are most likely do the same in both of them...

Mia_Material Advanced is a nice Shader, I love it. It's my main Arch Shader.

[Falam]

I actually also had a long winded post about the things i love about mental ray and the things i hate

I hope my previous moderating post wasn't the one to convince you to hold back. There is a fine line between heated debate and something ugly. I interjected, not because I thought this line had already been crossed, but because I wanted to prevent it from doing so.
"Long winded posts" are more than welcome, as long as everybody keeps it cool...
(me included, I hasten to add )

Who cares?

The OP for one and probably all the other users who have posted here so far... ;)

Nah your post had nothing to do with it, i was still (re)typing mine when you posted it.

And i pretty much stand by what i said, but just in case anyone truly cares:
what i love about mental is all the npr capabilities (if you know me that's no surprise, i go on about that stuff all the time ) and yeah the thing a lot of people seem to hate: the way you can edit and fiddle with nearly every aspect of it.

What i hate about it was pretty much well covered by Max ;) Also yeah Nizar displacement is pretty much a nightmare, but i found that annoying in vray as well..

But i still think my previous answer to Nizar is what matters - find a method that works best for you and refine it! Don't get bogged down in "ooh that other renderer has feature x and mine hasn't!" Unless it's Arnold of course

I have to re-read page #4 it was big, almost like a political debate
After reading some of it, it comes down to something I could have told you, Softimage has flaws, they are lax in areas, especially with Mental Ray that they need to focus on. With all that in mind, if I go with an alternative render I need to find the money tree seed and grow another money tree, so I can purchase another render, I think the seed was last seen in the Himalayan mountains (sarcasm) I'm not going to do this, so I guess the only next option for me and others is to Overclock your CPU