Why is everybody suddenly able to render using GI?

[sonne]

I was just wondering, if Mental Ray is the slowest GI renderer in the world?
GI-rendered animations pop up everywhere, but they're not rendered with Mental Ray. Instead it's Arnold and FastRender, and this is often small companies that doesn't even have a renderfarm.
How is that possible?
Is it just me, who is bad at tuning Mental Ray?

A few links:
http://www.3dluvr.com/brucester/
http://www.rithuset.se/

Thanx,
Peter

[theallusionist]

Xsi 2.01
Mental Ray 3
Greased lightning

any questions?

a

[sonne]

Are you sure you replied to the right post?
Are Arnold and FastRender faster than MR, or is it just me? Try to check out the links I provided.
The people behind the productions on Brucester only have their workstations to render on - no renderfarm - and still they are able to render 30 sec.animations using Arnold. That's pretty impressive in my opinion.

Peter

[wolfinger]

well, I just created a scene which has both an (4x4 sample) area light w/ shadows, 100 000 Global Illumination photons bouncing around, and Final Gathering enabled. The final render quality is every bit as good as I need for this project.

The precalc time (for GI and FG maps) is about 3 minutes on a Duron 900MHz, and the actual render time (albeit at a low resolution) is under a minute. In TV res it would probably take about 5 minutes total. Good enough for me, especially on this ancient computer.

Now, with a dual AthlonXP 1800+ computer I could multiply the number of lights and still render high-quality frames in affordable times.

[atyss]

Bonjour Sonne,

Well I must admit that this new breed of renderer (Messiah:Render - formely Arnold-, finalRender, Brazil, Vray, etc) is frightening when it comes to GI. But there are some issues when considering their usage.

• Altough for you it might not make a difference, these renderer does not produce true Global Illumination. They only do "radiosity", wich is a subset of Global Illumination. Radiosity takes into accound only diffuse information. But the great thing is that there is less calculation and less or no noise in the rendering.

• Since mental ray is so well implemented and since quite a long time, it is production proven. With it you can almost render anything, with beautiful results. mental ray is considered one of the best raytracers, if not the best. Its strong implementation also means that the workflow is better.

• Therefore, the mental ray pool of users is large, and thus it is easy to find support. This is not the case with newer renderers.

• As I see things, most of these renderers are not production ready. You'll have to make serious tests with them to know how they behave in production: like rendering high polygon scenes, scenes with lots of reflections, scenes with tons of light sources. One of the best test is to render a scene that take place in a room, and the only light source comes from a small window near the roof.

• Finally, do these need lot of handling? With mental ray, you worry about settings that are integrated into the software. Once you know the mecanisms, it become easy. Can you say the same thing about those renderers?

Hope this helps
Salutations - Cheers
Bernard Lebel

[pnoizat2]

is there a way of just telling mental ray to render without anything beyond radiosity?
without a whole page of script?
and under these circumstances does it render much faster?

worth a shot.......

[atyss]

@@@@
is there a way of just telling mental ray to render without anything beyond radiosity?
without a whole page of script?
@@@@

I don't understand what you mean. Can you elaborate?

Bernard

[pnoizat2]

if radiosity is a subset of global illumination, and requires far less processing than global illumination.

can we tick radiosity only box?
thus saving on calculations, time, hardware etc, at the expence of quality accuracy.

I'm sure radiosity is adequate for all but the most demanding representations.
Te Anau

[atyss]

Ok I see what you mean!

Well, subset of GI does not mean "a part of". It means that radiosity is not calculated the same way. So the answer to your question is.... NO! You can't render just radiosity in mental ray.

Here is a text taken for a mailing list (XSI or Render List on this site) where someone explained the difference between Global Illumination and Radiosity. I'm sure that it will be more helpful than me:

"Global illumination means that the radiance of a point depends not
only on the local charasterictics, such as BRDF and surface
normal, and the light sources, but possibly on the radiance of all
surface points that are visible from here. This interdependence is
expressed by the rendering equation. Looking from another point of
view, global illumination computation takes into account all light
paths including emission, single reflection, double reflection, triple,
etc, while local illumination models compute only the effect of
emission and single reflection. Ray tracing is somewhat in
between local illumination and global illumination. It computes
multiple reflections only for the ideal (mirror like) reflection and
ideal refraction.

So what is radiosity? Radiosity is a special case of global
illumination where all surfaces are assumed to have only diffuse
reflection, but, for instance, no specular (i.e. Phong or Blinn-like)
reflection. The reason of this restriction is that diffuse surfaces have
the same radiance in all directions, thus the radiance function will
depend only on the surface point. In general global illumination the
radiance also depends on the direction, which makes the
representation and the computation more complicated.
However, the diffuse assumption is often too restrictive.

There is often a confusion in the terminology. Radiosity is often
said instead of global illumination, because early global
illumination algorithms used this restriction. In order to emphasize
the difference, we can also talk about "non-diffuse global
illumination".

Global illumination algorithms are usually classified as finite-
element methods and random walk methods. Finite-element
methods require the surfaces to be tessellated to small triangles,
but do not exhibit pixel noise. Radiosity algorithms usually fall into
this category. Random walk methods try to find discrete samples
of the light paths connecting the light sources to the eye via
reflections and refractions. These paths are generated randomly or
quasy randomly since random samples are dense.

Mentalray implements a special version of Monte-Carlo bi-
directional path tracing, called photon map, and is able to solve the
general, non-diffuse global illumination problem."

Hope this helps
Salutations - Cheers
Bernard Lebel

[sonne]

I thought I wanted to share this with you guys - it's a mail I received from Diego Gutierrez:

• Peter

---

..I read an interview where Henrik Wann Jensen (who worked for mental images adding his photon mapping GI technique to the mental ray engine), said that it was funny to see how "the most expensives packages had the worst implementations" of similar GI techniques. He didn't mention mental ray explicitly, but it was pretty obvious in the context of the interview that he meant that. He walked out of mental images when they tried to gain exclusive rights to his work, (work that had been published by him before as his PhD. thesis)
Mental ray 3.0 is supposed to be a lot faster now. Haven't tried it yet, though, so I can't confirm that...

---

[wolfinger]

but you can render GI and/or FG without rendering raytraced lighting in mental ray. Would you not call this radiosity then? It certainly is faster than rendering raytrace + GI /and/or FG, especially when you have many area lights with shadows enabled. Of course it's limited, like for example you can't get any specularity this way, and it's very CPU-intensive to get good quality (relatively) sharp shadows.

[pnoizat2]

thanks.
I was quite in touch with the radiosity approach but hadn't nailed down GI. That helps heaps.
Te Anau

[atyss]

If you want a nice site with many resources on Global Illumination and Radiosity go check this out:
www.illumenal.com

Salutations - Cheers
Bernard Lebel

[cheeko8080]

Just want to clear up some misconceptions about one of those new renderers, namely Brazil.

• Altough for you it might not make a difference, these
  renderer does not produce true Global Illumination.
  They only do "radiosity", wich is a subset of Global
  Illumination.

Brazil is not a radiosity renderer, it does global Illumination.

Therefore, the mental ray pool of users is large, and
thus it is easy to find support. This is not the case
with newer renderers.

Don't think I can agree there. Brazil and at least one other of the mentionned renderer have public betas of their work available, and there's a surprisingly large group of people playing around with the software and ready to give support via the web in chat areas and webforums. Of course, not all of these users are "professional", but a good number of them are and the support you speak of is very strong.

You'll have to make serious tests with them to know
how they behave in production: like rendering high
polygon scenes, scenes with lots of reflections,
scenes with tons of light sources.

Brazil has been through a lot of these tests and is being written with these situations in mind. In fact, the renderer has recieved some flack before because a scene with two spheres doesn't render as fast as other renderers, but when it comes time to do that 100 million poly scene, the real advantage of the system becomes obvious.

• Finally, do these need lot of handling? With mental
  ray, you worry about settings that are integrated into
  the software.

Brazil is totally integrated into 3dsmax, it's not a translator and connects quite seamlessly into the standard max pipeline.

All in all, MentalRay has been around for awhile, and has certainly tackled a great deal many of the problems that these new renders are in the process of working on. For example, lazy loadng of geometry, which Brazil hasn't got yet (although it's been designed with that in mind), and so while I agree for the time being mentalray is more advanced, I think it is unwise of you to discard this group of new renderers offhand.

• Neil

[atyss]

Bonjour Soulburn3d,

It's nice that you cleared things up, I really appreciate.

@@@@
, I think it is unwise of you to discard this group of new renderers offhand.
@@@@

The goal of my post was not to discard newer renderers, as I'm sure many of them have a future. If you take a closer look on some of my comments...

-"But the great thing is that there is less calculation and less or no noise in the rendering."

-"most of these renderers are not production ready.." please note here that I said MOST, not ALL.

-"Finally, do these need lot of handling? With mental ray, you worry about settings that are integrated into the software. Once you know the mecanisms, it become easy. Can you say the same thing about those renderers?" Here I point out the fact that you need to check how the renderer's workflow can fit into the pipeline. I was just asking a question, not making a statement. You can see in my post that I say to be careful, not to discard those products.

So what I want to say is I'm sorry if you interpreted my words this way. My post was also meant to point out the benefits of mental ray in a production environment with XSI. The truth is that when time comes to purchase a renderer (or any other kind of software) you have to ask yourself the right questions. In the end, if the renderer does indeed satisfies all the criteria (ease of use, implementation, output quality, etc) well I strongly encourage its purchasing.

On a final note, production is the real test for a renderer. This is only when used in production that you can see what the renderer can deliver. Like you said, mental ray has come a long way. Now it's time for newer renderers to do the same. Believe me, I truly hope that new renderers will rise in production. I consider competition to be healthy for the user.

Salutations - Cheers
Bernard Lebel

[cheeko8080]

So what I want to say is I'm sorry if you interpreted my words this way.

Ok, that's cool, the tone of your original post (if there is such a thing as tone in this cyber world) gave me the impression that you were not giving these renderers a chance, not to mention you had some bad information in there.

Your criteria are sound, you do need to consider these things before deciding whether to puchase a renderer. I will not speak for the other renderers since I don't have enough information to make a sound statement, but Brazil is being written as a production renderer, written by production artists, not just as a GI renderer. It's core is being written to support a lot of the bread and butter stuff like largedatasets, memory management, etc, stuff that doesn't sound all that exciting, but anyone from a production background knows that reliability and overall quality is the important stuff at the end of the day, not how many whiz-bang features you have on the list.

• Neil

[pomme]

is there a easier or simple way to explain GI, FG, radiosity? Which is faster, what are their MAJOR different. And still, how come seems like everybody using GI to render their seen. I see most of the stuff that ppl post are rendered using GI.

[atyss]

Simply put:

GI: Photon mapping that takes into account diffuse, specular and reflection of objects.
Radiosity: Non-photon mapping that calculates only diffuse.

Faster: Radiosity
This is a theoric answer. In production things might not go this way. It's possible that with some elements (heavy geometry, complex phenomenas) a radiosity renderer becomes slower than a GI one. It's different with each renderer.

Major differences: GI produces noise, radiosity doesn't (or almost). GI more physically accurate.

@@@@
And still, how come seems like everybody using GI to render their seen.
@@@@

Because with mental ray you can produce GI and it's an easy way to light a scene. I suspect that hardware performance is a factor: GI has been there for a long time, however it's long to render. With faster machines it's now an affordable solution. I also suspect that the profusion of cracks allow much more user to benefit from GI renderers.

It's definetely not all users that use GI, because it is still very expansive in rendering times. It's just that GI produces beautiful images, and these pictures receive great visibility. Note that with strong lighting skills and a good renderer you don't need GI.

Salutations - Cheers
Bernard Lebel

[cheeko8080]

GI: Photon mapping that takes into account diffuse,
specular and reflection of objects.

Photonmapping is one way to simulate GI, however, you do not need to do photon maps in order to do GI. At least in general, I don't use mentalray and so don't know enough about their implementation.

• Neil

[atyss]

Photon mapping is only a way of calculation. There are others.

Bernard