Comparative render speeds.

[Cypher666]

Well technically if you wanted to use good GI an SSS you would have to use mental ray, which is the same for both programs (there may be some small difference but not much). One advantage of Maya is that you can also get renderman which is used extensively in the film industry, however it is usually quite expensive to buy a renderman (there are many) but there are a few excelent free rendermans out there too, 3delight is a good example:bounce:! I would try them both first and decide which one suits you best as it really is a preference thing:thumbsup:.

[Vozzz]

hm... ok cool thanks.

So mental ray is the way to go huh...

do you know if i can get a trial of it or something?

[Cypher666]

do you know if i can get a trial of it or something?

Mental ray comes built in to both maya and max these days, I assume that the trails of both of these programs have it already.

[Mauritius]

Well technically if you wanted to use good GI an SSS you would have to use mental ray

Define "good". And no, you would not have to use MenatlDelay. That's b**it. Sorry.

.mm

[Mauritius]

Sorry for the double post.

.mm

[Ian Jones]

Gotta chime in here for a sec, Mauritius... I don't think that was a fair call 'mentaldelay'. I know you have strong opinions, and valuable insight and help at times... but regardless of technical merit various rendering systems have various advantages and disadvantages. Mental Ray just so happens to be pretty easily accessible to most.

[Mauritius]

Gotta chime in here for a sec, Mauritius... I don't think that was a fair call 'mentaldelay'. I know you have strong opinions, and valuable insight and help at times... but regardless of technical merit various rendering systems have various advantages and disadvantages. Mental Ray just so happens to be pretty easily accessible to most.

The delay is about rendering speed. And I didn't invent that term. It was the 'official' (as what people that experienced the pain of its slowness) tag people in the industry in Germany used in the 90's. The matter of the fact remains that everyone cooks with water. I have yet to see a renderer that has ray-tracing as fast as MentalRay's can be but the image quality of a REYES renderer. I had to use MentalDelay on many projects when I worked in commercials and knowing the alternatives this product fails to impress me on so many levels that I gladly stick with calling it MentalDelay.

Apart from that I would also argue the 'easy' accessibility. It's an illusion. If you mask 80% of the useful features (in a high-end VFX context) of a product of MentalRay's complexity in a GUI wrapping it (e.g. Maya or XSI), it's not hard to make it 'easily' accessible. 'Easy' translates to 'weak' here unfortunately (where Maya's integration is really bad and XSI's is good only for the tasks boutiques or individual artists have to carry out as far as rendering goes). So seeing that as an advance is rather dubious. That's why people write plug-ins like this (http://sourceforge.net/projects/mrliquid).

And don't get me wrong. I think MentalRay is a great renderer. But it's far from being half as good as some people perceive it, who have zero experience in pushing renderers like that to the limit in blockbuster VFX. And it took people at MentalImages ages to understand some core concepts of high-end pipelines if you ask me. Only a few years back, Thomas Driemeyer from MI, in a post to the Highend3D Rendering Theory mailing list, regarded an ASCII-based scene format (such as e.g. RIB) as bullshit (he used more decent way of putting it, I recall) and sending data directly to the renderer (as XSI does with MR) as the only way. That shows total ignorance of the way high-end pipelines work. And now that they understood that, some overdue features have made it into the .mi format.
The decision to use renderer A or B fro big studios today is least driven by how well it is integrated into a package from an artist's pov or how fast its ray-tracing is. Both are featues that don't make much of a difference for those places. Other things are much more important.

The main reason imho, why MentalRay is used [more] in places that were more or less PRMan-only a few years back is that Pixar failed to deliver on the GI/ray-tracing side of things until recently. PRMan 13 luckily is a big leap forward in that regard.
However, I use 3Delight most of the time. These guys had a raytracing/REYES hybrid renderer in 2000. They were a few years ahead of Pixar in that area and it shows. Fast ray-tracing plus the image quality of REYES. But their product has gone under the radar of a lot of people.
And I'm a rendering TD. 'Easily accessible' for me means 'well documeneted'. Understanding how a renderer works internally I can use it properly.
Pushing buttons may make me feel I'm in control of things but the fact of the matter is that being able to drive a car and being able to repair it when it breaks down are two utterly different things. The former only gets you that far...

Cheers,

Moritz

[jbelleisle]

I've never rendered for the big screen--all my digital effects work has been for DV. I mainly use Vray. I take it you do big screen work, Moritz. What resolution do you need for that?

Anyhow, I hadn't heard of 3Delight before, but I checked out their gallery. The dog renders were very impressive. I started wondering--how does it do with other types of subjects? I understand that it mainly uses REYES, but how is its raytracing? Can it stand up to that of Vray, Brazil, or the like? One thing's for sure: that's the best fur I've seen.

I'm just taking a guess--is the main advantage of REYES that it can render at big-screen resolutions quickly and accurately? And maybe fur, too?

[Ian Jones]

Some great points, I've learnt a couple of things. Although having greater control, more under the hood flexibility is of course great... for many generalists who simply don't have the time and/or need to have that kind of control the software GUI's still make things accessible / easier. High end work affords more significance on specialists like yourself, where of course a GUI isn't neccesarily a useful interface or powerful enough tool. I just thought it odd that you had phrased it 'mentaldelay' but I understand moreso now. We were missing the context which is why I didn't think it fair to have such a statement without explanation... thx for talking more about what you meant.

[stew]

Anyhow, I hadn't heard of 3Delight before, but I checked out their gallery. The dog renders were very impressive.
I think such renders don't show off the capabilites of a renderer as much as they do show the skills of the person(s) who created that image.

I'm just taking a guess--is the main advantage of REYES that it can render at big-screen resolutions quickly and accurately? And maybe fur, too?
No, not necessarily. The maximum output resolution does not depend much on the rendering algorithm but rather on whether the renderer needs to keep the entire image around or not (which is why "good" renderers render in buckets and write to tiled image formats).

The Reyes advantages over non-Reyes renderers is that you always get pixel-perfect surface tesselation, minimal overhead for pixel-perfect displacements, fast depth of field and motion blur. Decoupling shading and hiding also gives you cheaper antialiasing withtout having to shade every single sample. Those are the advantages I can think of from the top of my head now.

[mr Bob]

Mauritius (member.php?u=470) I have to agree with you I was an ardent fan of MR , that is until Prman, which I find a dream to work with.

B

[joconnell]

Heya Vozzz,

I think an important thing to ask is what are you going to be using it for too? As mauritius states he really likes 3delight as it gives him a tonne of flexibility and he knows how he can get at all of the features thus he's gaining a lot of speed in terms of how quickly he can get shots to look the way he wants and set them rendering. He's working on really high end film work with a tonne of polygons in a big film pipeline that's probably got 3delight really well integrated with other scripts / programs that make it an efficient way of doing things.

Personally my situation is different. I'm using max which doesnt have a great pipeline to renderman but that said I work in a small post company that has 4 3d ops - we work on commercials that have quick turnarounds and we're all generalists. Our company has no rendering tds as for starters, we don't need them - our work doesn't justify having a shader writer or a programmer to make custom shaders - the jobs we do don't have the time or the budget to work that way. Because of tha I'm using vray which allows me all of the quality points of a renderman renderer and is incredibly quick for stuff like GI which makes my life much quicker for vfx work.

It's kind of a difficult question to answer without looking at what you want to do with the renderer - what type of stuff are you working on, what are the features you need - how much flexibility do you need in your shaders and passes, will you have a programmer to do this for you? Do you want to be a full time render td or do you want something that has enough features and a useable gui to get the results you need?

[Mauritius]

...really high end film work with a tonne of polygons in a big film pipeline that's probably got 3delight really well integrated with other scripts / programs that make it an efficient way of doing things.

We don't use polygons at all. We use subdivision surfaces or NURB patches. Polygons are evil and need to be destroyed(tm). They are a bad choice as an input geometry for a REYES renderer (or any renderer that is written to produce film quality images, I would argue).
Read why (http://www.plastickitten.net/wordpress/essays/reyes-primitives-some-philosophy/).
Did you know that neither the Terminator II nor the Dinsoaurs in Jurassic Park even were polygons? All NURB patches. One of the reasons ILM CG division developed REYES was that they relaized exactly that: Polygons are evil and need to be destroyed(tm).

.mm

[joconnell]

Yeah vray has an implementation of this in its current version that uses the same technology as the sub pixel displacement. Again it mightnt be the exact same thing but it's based on the same principle of quality being driven by the pixels it occupies.

[thev]

We don't use polygons at all. We use subdivision surfaces or NURB patches. Polygons are evil and need to be destroyed(tm). They are a bad choice as an input geometry for a REYES renderer (or any renderer that is written to produce film quality images, I would argue).For film this is probably so, because the number of patches is relatively small. However, polygons still have their uses - try to build and render a whole city block out of NURBS patches based on actual blueprints... you'll get nearly as many patches as you would have polygons and they will be slower to process.

[rendermaniac]

For film this is probably so, because the number of patches is relatively small. However, polygons still have their uses - try to build and render a whole city block out of NURBS patches based on actual blueprints... you'll get nearly as many patches as you would have polygons and they will be slower to process.

You'd most likely build pretty much everything as subdivision surfaces these days if your renderer supports it. You get smooth surfaces, the renderer deals with tesselation (REYES at least) and joining surfaces is so much easier.

For characters there is no contest - subdivs every time.

Simon

[Mauritius]

For film this is probably so, because the number of patches is relatively small. However, polygons still have their uses - try to build and render a whole city block out of NURBS patches based on actual blueprints... you'll get nearly as many patches as you would have polygons and they will be slower to process.
You'd use bilinear patches where possible. Well defined normal, well defined intrepolation, built-in uvs.
Or you would use subdivision surfaces with creases. Well-defined normals, well defined interpolation, nice subtle highlights on edges & corners for free, better u-poly distrbution if you have to do e.g. ambient occlusion etc.
Plus more lightweight than the polygon mesh as you can do bevels using creases & corners vs. using actual geometry.
Nah' really, you don't need polygons for that anymore. :)

.mm

[Mauritius]

Yeah vray has an implementation of this in its current version that uses the same technology as the sub pixel displacement. Again it mightnt be the exact same thing but it's based on the same principle of quality being driven by the pixels it occupies.
This is not about sub pixel displacement. That is just a nice side effect of REYES. If you think my essay is about sub pixel displacement you totally missed the point. Plus we are talking stuff here that was available in other renderers since 1988.
I'm totally unimpressed by VRay, Brazil, etc. adding stuff like that in 200X...

.mm

[joconnell]

This is not about sub pixel displacement. That is just a nice side effect of REYES. If you think my essay is about sub pixel displacement you totally missed the point. Plus we are talking stuff here that was available in other renderers since 1988.
I'm totally unimpressed by VRay, Brazil, etc. adding stuff like that in 200X...

.mm

No I don't think that at all - what I'm saying is that vray has added in subdivision that are based on pixel area like what you describe in your essay so it will raise or lower the level of subdivision dependant on what screen area the object takes up. What i was saying is that this subdivision was introduced with their sub pixel displacement.

All i'm saying overall is that the best renderer for the job isn't always renderman - it's a question far too dependant on your current circumstances, level of knowledge, what your needs are, what your resources are and the type of work you are doing.

[silvia]

Is there any documentation out there about render time comparison between Renderman and Mental Ray? Sure Renderman is faster when the model complexity and displacement are the main concern, but how about ray tracing, global illumination, caustics, SSS, etc?
Also, does anybody know how Maxwell compares to the 2, both as quality and speed?
I work with MR right now and I think MentalDelay is really the best way of calling it. Lighting, shading and rendering by their very nature are a render-tweak-render-... process, and the quicker the render part, the more time I can dedicate to the lighting and shading parts. I am even willing to do a substantial money investment in another renderer, if it gives me better performance for similar quality.

[Mauritius]

Is there any documentation out there about render time comparison between Renderman and Mental Ray? Sure Renderman is faster when the model complexity and displacement are the main concern, but how about ray tracing, global illumination, caustics, SSS, etc?

No, because RenderMan (http://en.wikipedia.org/wiki/Renderman) is an abstract interface and MentalRay is a renderer. :)
You probably mean PRMan. Anyway, when people say 'RenderMan is slower than MRay' they often limit their view to PRMan.
But there are no comparisions I'm aware of. To create a comparable setup with renderers using that different approaches in their architecure can be fairly problematic.
There are also three other ray-tracing capable RenderMan-compliant renderers: 3Delight, AIR and Pixie. The former two are commercial, the latter is an OSS project.

3Delight and PRMan use very differnt approaches for ray-tracing a scene. 3Delight also allows compiling shaders to machine language which gives a big speed increase for certain shding problems.

I'm fairly certain that under really comparable circumstances as far as number of rays shot, number of polygons that rays intersect with and shaders being executed on ray hits goes, MentalRay and the more mature RenderMan compliant renderers aren't that far apart, if different at all. Everyone cooks with water and most people that state that a REYES/ray-tracing hybrid renderer is slower than a scanline/ray-tracing one like MentalRay, have either not verified this themselves ever (they just repeat what they heard other people say) or they have only looked at PRMan vs. MentalRay which is a quite narrow field of view.

3Delight is a REYES/rayracing hybrid renderer as well and it had ray-tracing years before PRMan and as I said, it uses a very different approach in how it preps geometry for 2ndary ray intersection tests.
Oh, and you can actually force 3Delight and Pixie to be full ray-tracers (aka camera "rays" are real "rays" and not pixel/sample/scanline intersections) by selecting a specific "Hider".

Cheers,

Moritz

[mr Bob]

Is there any documentation out there about render time comparison between Renderman and Mental Ray?

why does it always seem to be looked at as being black or white ? for starters PRman and other such renderers are not designed for home users / small shops who want a few buttons to push for a solution.If you have a few blade servers loaded with xeon's and gigs of ram coupled with some shader writing skills, it doesnt really matter weather this app or that app is faster out of the box because both PRman and MR are flexible enough for a firm to lever out whatever they want.

B

[silvia]

why does it always seem to be looked at as being black or white ? for starters PRman and other such renderers are not designed for home users / small shops who want a few buttons to push for a solution.If you have a few blade servers loaded with xeon's and gigs of ram coupled with some shader writing skills, it doesnt really matter weather this app or that app is faster out of the box because both PRman and MR are flexible enough for a firm to lever out whatever they want.

B

If you notice, from my post, I assumed that it is NOT black or white. I am pretty sure that the 2 have different strenghts in different situations. I was simply curious to know if there have been comprarisons using the same scenes or not. For example, since Maya plugs into both, I thought somebody may have had the idea of rendering the same scenes with each and see what happens. Maybe not...
I am not sure where you got the impression that I want a 1 button solution. I use Mental Ray: do you consider it a 1 button solution?
Also, exactly because big companies can afford all the computing power they need, and smaller users don't, the speed becomes even more crucial for the small users.

PS: by the way, sorry Mauritius, you are right, I was refering to PRman.

[ThirdEye]

I use Mental Ray: do you consider it a 1 button solution?

Unless you're writing your own MR shaders and compiling them i guess you're using a 1 button solution... well maybe 2 buttons, i'll give you that ;)

About the polygons vs everything else... Wouldn't the best thing to do be using polygons and converting them to renderman subds at rendertime?

[silvia]

LOL, ok I will settle for 2 buttons... ;)

[bahman]

If you have a few blade servers loaded with xeon's and gigs of ram coupled with some shader writing skills, it doesnt really matter weather this app or that app is faster out of the box because both PRman and MR are flexible enough for a firm to lever out whatever they want.

B

If i have it , I'm sure that Prman is the best renderer for me, becuse of simple shading language, very good documents, and stability in massive scenes.

[rendermaniac]

No, because RenderMan (http://en.wikipedia.org/wiki/Renderman) is an abstract interface and MentalRay is a renderer. :)



I'm sure I read somewhere that Mental Images .mi format has been presented as a standard - it's just that Mental Ray is the only renderer that implements it.

Simon

[Mauritius]

I'm sure I read somewhere that Mental Images .mi format has been presented as a standard - it's just that Mental Ray is the only renderer that implements it.
Yeah, and what happens if you implement a parser for that format in a commercial renderer ... guess...? :)

I know for certain that at least one RMan compliant renderer can already parse and render .mi files.
Still, that version is not available publicly so far. My guess would be that this is mainly because of the totally fuzzy legal situation...

At least, all RMan compliant renderers out there have a license from Pixar. Rumor has it that Pixar has stopped giving licenses to create RMan compliant renderers since a few years. If that is true, then RenderMan as a cross-renderer standard only exists as long as those other renderers exist.
Exluna has found a nice way around it with their parser plug-ins. Doesn't help with shaders though. And the fact that MentalDelay's "new" MetaSL is nothing but a copy of RSL with structs added (which is an overdue addition for RSL too) but different enough syntax to make sharing code a pain doesn't help either.
Fact of the matter is that commercial vendors of renderers don't see any adavance in supporting an open standard for scene description and shaders as it takes control of their markets away from them.
It is kinda funny that this is exactly the opposite pov of the one their users in the VFX industry have. But then VFX is a niche market for many vendors.


.mm

[silvia]

Of course, and they make even more money by developing their own (expensive) plugins into the various applications to do that format conversion for you.

[rendermaniac]

Exluna has found a nice way around it with their parser plug-ins. Doesn't help with shaders though. And the fact that MentalDelay's "new" MetaSL is nothing but a copy of RSL with structs added (which is an overdue addition for RSL too) but different enough syntax to make sharing code a pain doesn't help either.


I got the impression that MetaSL was more like GCC - a frontend which can compile to several languages. If I read it correctly then it would be possible for someone to write a RSL backend for it. I don't think you could do a lot of stuff MetaSL does in pure RSL anyway - it is designed for graphed shading networks - RSL would need to inline it all. If nothing else it could give Pixar a few ideas for RSL.


Fact of the matter is that commercial vendors of renderers don't see any adavance in supporting an open standard for scene description and shaders as it takes control of their markets away from them.
It is kinda funny that this is exactly the opposite pov of the one their users in the VFX industry have. But then VFX is a niche market for many vendors.

.mm

Yeah well I think they are only promoted as open standards so that people can write exporters to their format to get more people using their renderer.

I fear I have gone a little offtopic - sorry!

Simon

[bardakos]

I will hijack for a sec the thread but ...sicne the topic is Comparative render speed i'm not off topic...

For fast TV based work commercial,motion graphics etrs we found out that LW renderer is much better than MR in terms of setup/render time/license cost.

So we never climbed the MR train ...

We are in need of a better Renderer tho so we decided to try Mantra (Houdini's renderer)
but we have no info on speed comparing MR or Prman ,Air etc...

(since we decided to get Houdini nevertheless for vfx stuff)

any ideas or info about that?

is it worth to get one Houdini (Master)-->9K with current offer and add to it a few AIR licenses
or is it better to get one Houdini Escape and add to it extra Mantra Licenses...

..

[stew]

We are in need of a better Renderer tho so we decided to try Mantra (Houdini's renderer)
but we have no info on speed comparing MR or Prman ,Air etc...
"Better" has different meanings for different people. What are you looking for in a renderer and what are the parts about Lightwave's renderer that you are unhappy with?

[bardakos]

1.Lack of caching ...(other that surface baking )
2.Lack of Shadow Shading...
3.Our work has to be raytracing dependent and we would like to expand into a more "shading" based solution..so that our team gets the experience with such kind of ...trouble :)
4.Windows Systems only (for workable stability)
5.Displacement seems slow...and geometry needs to be pre subdivided in a really high level...


We do use LW and we will continue to cause it is effective in all means..
Specially with LW9 it seems much faster not only the setup thing but the actual rendering itself...
Not to mention that Fprime is a big asset..

Now for special projects we are about to confront we need to use a procedural workflow
and take advantage of Chops for channel manipulation through various audio inputs.So we will use Houdini..which comes with Mantra ..so... :-)

[2stepdmb]

At least, all RMan compliant renderers out there have a license from Pixar. Rumor has it that Pixar has stopped giving licenses to create RMan compliant renderers since a few years. If that is true, then RenderMan as a cross-renderer standard only exists as long as those other renderers exist.
.mm

Actually, a written license is no longer required. Quoting Chris Ford from a post on the Pixar's RenderMan forums:

"This is notification that a written license from Pixar is no longer required for any software products and tools written to the RenderMan Interface Specification (RISpec). The standard specification concerning Pixar's copyrights and trademark rights has now been updated to permit use of the interface by other software applications as long as Pixar's copyright is acknowledged. Please refer to the revised specification for more details.

https://renderman.pixar.com/products/rispec/index.htm"

[d7man2000]

my guess is to learn one program well, to model organic/nonorganic everything

when time comes, it all comes to basics, modeling, texturing, lighting, paying attention to details and having a lot of patience when program crashes and file that you just saved just got corrupted and deadline is in 5 minutes

people writing about renderman and all those expensive additions must spend a lot of time learning all this stuff with shaders and scripts, truth is that small percentage of people really get through knowing this stuff in a real environment

for fun you can learn anything,
when it comes to a moment when you get old enough and still don't create graphics, situation changes and you wake up realizing that most stuff that you know is not required by many places and there will always be better people in everything already knowing this stuff and they are a lot younger, that's really true
when you hope for a dream job in a studio making another Shrek, better wake up sooner than later
i started with autocad, then bryce, then cinema, lightwave, maya, 3ds max, i even finished classes for maya, i learned 3ds max on my own with a student version of the program that i purchased, i think it was a wise decision. i still use maya helping my friend with his deadlines but it happened only twice so i can't say this lifestyle can feed you well

learn to make humans, you will always be needed wherever you go, if you master that, you have nothing else to prove, you will become another artist.

[Jozvex]

Is there any documentation out there about render time comparison between Renderman and Mental Ray? Sure Renderman is faster when the model complexity and displacement are the main concern, but how about ray tracing, global illumination, caustics, SSS, etc?

Check out this thread in the Maya > Rendering forum:

http://forums.cgsociety.org/showthread.php?t=371816

You'll have to ignore the initial bickering, but after that it has some great info in it.

Also, does anybody know how Maxwell compares to the 2, both as quality and speed?

Well, I have used all three renderers (though PRMan the least, I used to use BMRT for a while though), and currently own MentalRay (through Maya 7) and Maxwell Render. In terms of 'features of a production renderer for animation/vfx work' Maxwell has almost none.

The current slogan for Maxwell is 'as easy as taking a photo', and that's fairly accurate. So for any kind of animation work you'd be taking 5000 (or whatever) individual photos rather than simply rendering out an animation. It has no animation capabilities so far apart from using it's Emixer feature to do relighting type animations. Some animation for the Disney Channel has been done using Maxwell by Keytoon Studios and it looks great, but they were very well equipped and prepared for the challenge of using Maxwell for animation in it's current form.

What CAN Maxwell do? Render polygons. That's about it! No NURBS or SubD or fur/hair, or renderpasses or baking etc like RMan and MR.

Speed? That's a largely debateable topic but basically Maxwell is very slow. It's debateable because Maxwell is unbiased and also has every bell and whistle being calculated all the time whereas for RMan or MR you only use the specific features the image/pass requires, and they're biased. If you rendered an image in a RMan renderer (or MR etc) in the most unbiased way possible using every form of light interaction they could, they would probably be slower than Maxwell.

Quality? I would have to say that Maxwell's quality is extremely high...given enough rendertime. An image you render starts as a mass of noise, but after only 2 minutes (depending on the scene I guess) you can see a fairly accurate image of what you'll get, but it'll still be really noisy. The longer you leave it rendering the cleaner the image gets. Each version of Maxwell is getting a bit faster and less noisy for sure, which is nice.

When you render in Maxwell, you're getting:

* Everything is high dynamic range
* Direct light, indirect light, direct caustics, indirect caustics, true dispersion, SSS etc etc
* Accurate lens effects like DoF, bokeh, lens diffraction (glare/lens flares), vignetting, distortion etc
* Spectral calculation rather than just RGB
* Normal mapping and bump mapping

And some other stuff I'm forgetting. You can adjust the lighting in realtime as you render or after the render is finished. It has network rendering etc.

One of the greatest things about Maxwell is that it's based on real life and so all it's units, camera settings, lighting etc is all using the real-world values. You can recreate a studio setup and plug in the same lights' strength in Watts/Lumen/Kevlin/etc, use the same fstop/iso/shutterspeed etc and get the right result.

So basically it's really quite hard to compare Maxwell to regular renderers. For realistic stills I love it.

EDIT: Oh, I forgot to even mention that Maxwell has about 10 renderer settings. MR probably has over 100 just displayed in the Maya UI alone. Also, Maxwell really only has 1 material that does everything. No need to link up 20 nodes all achieving different effects like fresnel reflections, glossiness, photon shading etc. The Maxwell material also has extra effects like thin-film interference and the ability to load measured BSDF and IOR files for super-accuracy.

I should also say that the main reason I'm writing all of this stuff about Maxwell is because it's new and not that many people around here have experience with it.

I do most of my work with Mental Ray though and like it equally as much as Maxwell for all the opposite reasons like flexibility and total creative control etc.

[rendermaniac]

1.Lack of caching ...(other that surface baking )
2.Lack of Shadow Shading...
3.Our work has to be raytracing dependent and we would like to expand into a more "shading" based solution..so that our team gets the experience with such kind of ...trouble :)
4.Windows Systems only (for workable stability)
5.Displacement seems slow...and geometry needs to be pre subdivided in a really high level...


We do use LW and we will continue to cause it is effective in all means..
Specially with LW9 it seems much faster not only the setup thing but the actual rendering itself...
Not to mention that Fprime is a big asset..

Now for special projects we are about to confront we need to use a procedural workflow
and take advantage of Chops for channel manipulation through various audio inputs.So we will use Houdini..which comes with Mantra ..so... :-)

THe thing I would think about is how you are going to get your data into Mantra.

Are you going to be using Houdini for all your modelling/animating/texturing too or will this be done in another program?

I don't think importing into Houdini is that hard. But I also don't know of any exporters for Lightwave (granted I can't think of any production ready RenderMan exporters for LW either!).

If you use Maya as well the answer is easy as there are at least 3 ways of getting data into a RenderMan complient renderer (MTOR, RfM, Liquid).

Simon

[Freak!!]

I don't think importing into Houdini is that hard. But I also don't know of any exporters for Lightwave (granted I can't think of any production ready RenderMan exporters for LW either!).
There are 3 RIB exporters for LW i'm aware of. Objects/textures are also easy enough to get into Houdini as well.

I still think AIR maybe a better option on the rendering side, than Mantra.

In terms of answering the original posters question.
Pure GI speed i'm guessing commercial renderers like Kray and Vray for GI speed.
MR likely has a good balance between the GI and SSS options, but when the next FPrime is released, SSS speed and GI will both be good in FPrime too.

[silvia]

Thank you Jozvex, very interesting!

[rendermaniac]

There are 3 RIB exporters for LW i'm aware of. Objects/textures are also easy enough to get into Houdini as well.


Out of interest - what are they?

Simon

[Freak!!]

Out of interest - what are they?

Simon

You can find the links for three of the free ones from here:
http://www.renderman.org/RMR/OtherLinks/index.html

Lightman -

LtoR -

Light-R -

Iv'e used two of them with varying succsess before.....
And there are also two slightly better commerical ones too...

PS. Polytrans also can convert Lightwave to .RIB.

[billrobertson42]

Did you know that neither the Terminator II nor the Dinsoaurs in Jurassic Park even were polygons?

Perhaps had more to do with the technology and computing cost/power of the time. T2 was done 17 years ago, and Jurassic Park was made 14 years ago (damn, getting old).

The rise of polygons and sub-d's has only been in the last 6-7 years right?

[stew]

The rise of polygons and sub-d's has only been in the last 6-7 years right?
Bah! They used polygons in Tron. :D

Catmull-Clark surfaces were AFAIK introduced in 1978(!) (link to publication (http://www.idi.ntnu.no/~fredrior/files/Recursively%20generated%20B-spline%20surfaces%20on%20arbitrary%20topological%20surfaces.PDF)), but I don't think they've been used much until ~20 years later - Geri's Game is from 1997.

[CaptainObvious]

This is not about sub pixel displacement. That is just a nice side effect of REYES. If you think my essay is about sub pixel displacement you totally missed the point. Plus we are talking stuff here that was available in other renderers since 1988.
I'm totally unimpressed by VRay, Brazil, etc. adding stuff like that in 200X...

.mm
On the other hand, Vray is probably a better choice than Prman if you need to render complex global illumination quickly. ;)


What CAN Maxwell do? Render polygons. That's about it! No NURBS or SubD or fur/hair, or renderpasses or baking etc like RMan and MR.
Not all polygons, either! Maxwell only renders triangles, I think. Or was it triangles and quads?



Anyway, I've gotten some really amazing render times in modo 201's renderer. I once rendered 1.6 million polygons (diced up subdivision surfaces) in less than two minutes, using multiple bounces of high-quality GI, and that was on a single 1.6GHz G5. However, I don't think modo's renderer is there just yet. It has quality and speed issues when using irradiance caching, and using non-cached GI is just too darn slow. However, most of the speed issues are to be fixed for 202 which will be announced on Siggraph.

I've also gotten amazing results pretty quickly in Kray (http://www.kraytracing.com/). In a certain sense, it's a Vray-like renderer for Lightwave (ie, dirty little tricks all over the place to keep render times down, and it's great for arch viz). But there are loads of parameters to tweak in Kray to keep the render times good. Screw them up, and you'll be rendering for days... But I suggest keeping an eye out for Kray 1.7.

[jude3d]

I think mental ray is a great raytracing solution, fast and complete rendering.
Renderman is the most amazing render for multi-pass rendering I think, really fast for every situation and especially for big scene.( I like a lot 3dlight and i was pretty imprest by rman 13.0, amazing new capabilities in raytracing and speed.)
renderman is really complicate for beginners, but Rfm 2.0 and the next renderman studio will be totaly integrate with maya interface, so it will be a nice solution for everybody. The power of a such renderer is in the custom scripting and shader programming of course but it's like to compare a ferrari and a commercial car, both could give amazing result if you are pilot, but you could do much better with the ferrari. renderman has no limitation in rendering.

one last raytracing rendering is modo 2.02, this is for me one of the best and fast rendering I ever seen, great GI, SSS, and really fast. I love to play with .

you must think about a renderer with the work you need to achieve. If you have a big scene with milions of polygon and displacement and not a lot of time , renderman will be the best.
if you just want to need style image or simple Gi animation, mental ray or modo will be great.
you need to think what result you want and wich amount of time you got to do the job , and that's it.

If your not a professional or TD don't use Rman for instance. try MR, Vray or modo 2.02

[WireFX]

Hi all!

It´s amazing, the thread has has the title "comperative render speeds" and we are currently again at the dicussion about the usability paradigm of renderers without ever having posted some sort of benchmarks ... what leads me to the speculation that the renderspeed is negotiable in comparison to the usability of a renderer to get satisfying results.

I am currently supporting a bunch of design students at a design university in germany and one thing that is really interesting is that most of them simply don´t care about rendertimes. They really deny from their inner desires to want to understand how a renderer works. I have seen here toon renderings which took about 7 hours, and the student got really astonished as I told him that he don´t need a reflection depht of 7 in a toon rendering without reflections, and that there is a different renderer in Max called Mental Ray, and that there is a lot of stuff he can disable when rendering only toon style. Well but as I said they don´t care very much ....

So for me there are 2 usergroups, the ones that are simply taking the renderer you installed on their machines and make all settings as high as possibel in some sort of "fire & forget" manner ... and the second group which is getting annoid when they have to use a software that restricts them and doesnt allow them to use a voronoi-noise in conjuction with a Blinn-highlight. The last ones sooner or later make their way to renderman ... ( or "mentalray-highend-knowledge", or gelato, or maybe (when they make it renderman compliant) Brazil ..)

[CaptainObvious]

It´s amazing, the thread has has the title "comperative render speeds" and we are currently again at the dicussion about the usability paradigm of renderers without ever having posted some sort of benchmarks ... what leads me to the speculation that the renderspeed is negotiable in comparison to the usability of a renderer to get satisfying results.
In some renderers, you have total control over almost all parameters used for rendering. This lets you optimize them on a scene-specific level for better rendering quality and speed. Of course, it also makes them harder to use than the ones where you just set up a couple of parameters and hit "go." Vray is a fast renderer, yes, but Vray with poor settings is going to render much slower than a renderer that is technically slower. Turn up all the settings in Vray to the maximum, and you'll probably render more slowly than Maxwell. ;) Other renderers, like Fprime or Maxwell, hardly lets you control anything. This makes them very straightforward to use, but it makes it impossible to optimize them as much.

Measuring the speed of a renderer is not just the matter of figuring out how many ray/polygon tests it can perform in a second on a given scene and a given hardware platform. There's much more to it than that. If I spend five hours tweaking my render settings, and then render in half an hour, that's still more time than if I spend five minutes tweaking my render settings and then render to the same quality in four hours. Unless, of course, I have to render a whole bunch of frames...

[mocaw]

...I spend five minutes tweaking my render settings and then render to the same quality in four hours. Unless, of course, I have to render a whole bunch of frames...

That's very true. Depending on who you are, and what you do, looking at how you get from point a to z is important. Rendering is only part of it- though a very important one, but over all man hours spent on a project + computing time is the real bottom line.

[billrobertson42]

This lets you optimize them on a scene-specific level for better rendering quality and speed.

Do any of them give you some sort of profile on where they're spending their time so you can optimize efficiently? Or are you just left to guess at what could be makint it slow?

[rendermaniac]

Do any of them give you some sort of profile on where they're spending their time so you can optimize efficiently? Or are you just left to guess at what could be makint it slow?

Prman 13 gives you several pages of XML statistics which show you pretty much everything you'd want to know - render time, micropolygon dicing, ray hits, texture accesses etc. It also has a CPUtime pass which colour codes the image so you can see which bits take longer to render.

They have also added a shader profiler which breaks down the time spent shading pretty much line by line per shader if you want to!

I haven't used Mental Ray as much, but I would not be suprised if it gives you a similar amount of statistics. Plus shaders are C so I am sure they could be run through a profiler if you wanted to.

Mental Mill looks like it could come with some visual debugging tools. Prman doesn't have this yet, but this is more useful for shader writers than speeding up renders.

Simon

[CaptainObvious]

Prman 13 gives you several pages of XML statistics which show you pretty much everything you'd want to know - render time, micropolygon dicing, ray hits, texture accesses etc. It also has a CPUtime pass which colour codes the image so you can see which bits take longer to render.

They have also added a shader profiler which breaks down the time spent shading pretty much line by line per shader if you want to!
That sounds pretty damned awesome.



Hmmm. I wonder if it's possible to use Apple's Shark or some other such performance-optimization tool to find out which takes the most time...

[Cypher666]

Define "good". And no, you would not have to use MenatlDelay. That's b**it. Sorry.

.mm

yes that was bad phrasing. Don't get me wrong, I love what renderman can do, however as an artist it can be very daunting to begin with, I read like 2 books on renderman and through it in the too hard basket (prolly cause I'm a max user and am still learning maya). It won't be long until I pick up on renderman and, like I usually do with new software, will never look back.

P.S. I see you're over at Rising sun, maybe you could give me a few 3delight lessons one day LOL:D

[jiversen]

We are in need of a better Renderer tho so we decided to try Mantra (Houdini's renderer)
but we have no info on speed comparing MR or Prman ,Air etc...

(since we decided to get Houdini nevertheless for vfx stuff)

any ideas or info about that?

is it worth to get one Houdini (Master)-->9K with current offer and add to it a few AIR licenses
or is it better to get one Houdini Escape and add to it extra Mantra Licenses...

..

Hi there,

I make periodic comparisons on several features with Mantra versus PrMan - which is widely regarded as THE defacto standard for a high-quality, well-performing renderer and so here is a thread where I test some of the easier-to-test features; ambient occlusion, GI, motion-blur and DOF. It doesn't go deeply into big scenes with large production textures, deep shadows and such, but it was something I could do in half a day of overhead time.

http://odforce.net/forum/index.php?showtopic=3555

I'm a fan of Mantra, and it becomes very cost effective because they have a really good licensing scheme for customers with several licenses. It might be possible to end up with a site license if you persue this route; I'd just speak with them and find out the best way to go with it.

[CaptainObvious]

Interesting test of Mantra vs PRman, Jiversen. Now, if I may be so bold: in modo, that scene would use less than 200 megabytes of memory (including bucket and frame buffers). I don't know what hardware it was rendered on, so I can't really comment on the render times, but chances are the render times would be less. I've rendered larger images than that made up of millions of polygons with full GI in less than five minutes, on low-end hardware.


Edit: No, I'm not saying modo's renderer is better than Mantra or PRman. They're not really comparable, as such.

[jiversen]

Interesting test of Mantra vs PRman, Jiversen. Now, if I may be so bold: in modo, that scene would use less than 200 megabytes of memory...
Edit: No, I'm not saying modo's renderer is better than Mantra or PRman. They're not really comparable, as such.

Well, that is a bit bold - (and, for sure you may be correct since Modo has specialized renderer and not so much a general renderer) - but it doesn't seem right just to say that without actually having this geometry and doing the equivilant test with it. Not to be a stickler and a killjoy but we probably can do harm to Modos reputation with unsubstantiated claims.

[CaptainObvious]

That's why I said we can't really compare the render times, you see. :) I don't mean to make any unsubstantiated claims. You wouldn't happen to be able to share that particular scene, would you? It would make an excellent testing base.

[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.