Max - GI - Animation... Which render engine?

[Ooze3d]

By the way... Is it worth setting up a little render farm with my desktop (P4 3Ghz - 2Gb RAM) and my laptop (Centrino 2Ghz - 1Gb RAM)? Or it will be just sligthly faster?

Thanks.

[Myliobatidae]

I would recommend Vray, it has some great time saving features, as well as being pretty fast by itself, however these features really only work when the camera is moving, not other objects, in the event you have other objects moving I think you would have to use brute force GI, when its just the camera moving you can save you direct light source to a file and reuse it, the same goes for your indirect, so that can save alot of time, also using vray proxy meshes in place of high poly objects can really save you time and memory, and vray displacement works really well also, if anyone using finalrender wants to jump in and talk about that, I would also like to hear some of its features, as for Brazil you'll have to wait for version 2.0, their current version cannot compete in the speed area...

[Myliobatidae]

By the way... Is it worth setting up a little render farm with my desktop (P4 3Ghz - 2Gb RAM) and my laptop (Centrino 2Ghz - 1Gb RAM)? Or it will be just sligthly faster?

Thanks.

Yes, it will be atleast 33% faster maybe more...

[Ooze3d]

Great! Thanks for the answers, Myliobatidae!

Another opinion from a FinalRender user...? Or maybe even Mental Ray?

Antoher thing that comes to my mind when trying to achieve a nice and realistic result is Motion Blur. How good is Vray's? Is it going to add much more rendering time? What about the other renderers?

Thanks again.

[ignanacho]

I think you can do it with Vray

Using GI. but saving your irradiance map and reusing it

First you do a render every 20 frames to save the irradiance pass and the you reuse it on your final animation. That saves you a lot of time

If you want me to explain better just say so (maybe you already know how to do it)

Ignacio

[Ooze3d]

I think I understand (it's not my first contact with GI) but if you could explain it a bit more or point me to a tutorial it'd be great.

Thanks!

[alejandro_m1]

For rendering still enviroments with camera movement what has been told about precalculating irradiance map is the best you can do, It will be the quickest and with most quality (specially less flickering) than with other methods. A tutorial can be find in vray help here:

http://www.spot3d.com/vray/help/VRayHelp150beta/tutorials_imap2.htm

For the ones with moving objects you are pretty much out of luck, unless you composite them which will be the fastest. When using moving objects you can´t pre calc irradiance or light cache maps, so it must be calculated each frame, also irradiance tends to flicker a lot if you don´t adjust everything correctly (QMC samples, IRR HSph and interp samples, blur gi among others), this also tends to send render times rocket high. QMC tends to be less flickery but really noisy which usually can be cleaned easier in a post app.

For moving objects I would really recommend following this tutorial

http://www.spot3d.com/vray/help/VRayHelp150beta/tutorials_anim.htm

[Ooze3d]

Gracias Alejandro!

Thank you very much!

[Myliobatidae]

Great! Thanks for the answers, Myliobatidae!

Another opinion from a FinalRender user...? Or maybe even Mental Ray?

Antoher thing that comes to my mind when trying to achieve a nice and realistic result is Motion Blur. How good is Vray's? Is it going to add much more rendering time? What about the other renderers?

Thanks again.

Here's where it get interesting, there is always a trade off, a reyes renderer will fly through motion blur probably as fast as vray without motion blur, the same holds true of displacements, but reyes renderers are slow when it comes to GI...

By the way the reyes renderers are renderman compliant renderers such as Aqsis, 3Delight, Air, and of coarse Prman...

There might be another option though, they just released a free version of Gelato, and I was reading somewhere on here that it does motion blur very fast, you might wanna check that out...

[Ooze3d]

I did check gelato out and they say it's quite fast for both motion blur and gi BUT you have to own a NVIDIA graphic card cause it's the first (that's what they say) engine using a 3d accelerator chip to render. My problem is that I've got a Radeon card.:sad:

Does vray get much slower when adding motion blur?

[alejandro_m1]

It does get slow, being out there solutions that do this a lot faster, but if you are following the idea of compositing later you could save the velocity channel and use it to have some great motion blur in Combustion/Fusion/Shake. I actually prefer this method since you have more control on how much motion blur you want to add and if you don´t like the way it looks you don´t have to render again everything, just tweak some parameters. I actually never use the camera effects (motion blur, depht of field) in vray, I just prefer doing all that in post.

Espero que te ayuden estas ideas, saludos!

[Myliobatidae]

I did check gelato out and they say it's quite fast for both motion blur and gi BUT you have to own a NVIDIA graphic card cause it's the first (that's what they say) engine using a 3d accelerator chip to render. My problem is that I've got a Radeon card.:sad:

Does vray get much slower when adding motion blur?

Well the verdict is still out on weather or not Gelato is of any use, but I am able to use it, using just a $55.00 graphics card, it just has to be above a Gforce 5500 I think, I have a 6500, so its not like you have to go out and buy a $1,000 graphics card...

with that said, we'll soon find out...

With vray your render times will probably quadruple, there is always the option of doing it in post, not as acurate, but it could be acceptable...

[Nii]

On the topic of mini render farms, would an extra computer compute alternate frames, or actually process parts of the same frame?

[alejandro_m1]

You can configure it to do parts of the same frame or rendering each one a different frame which is the adviced one, it´s faster, it´s safer and you don´t have composition problems later, the only case where rendering a single frame is adviced (like Vray´s distribute rendering) is when you have a single and really big still with tons of detail.

It´s also adviced, if you are rendering animation, to save the render as an image sequence, that way if one of the render nodes crashes or there is a power failure you can recover your work from the last rendered frame, instead of having a .mov, .mpg or an .avi that become corrupted.

[Mauritius]

Here's where it get interesting, there is always a trade off, a reyes renderer will fly through motion blur probably as fast as vray without motion blur, the same holds true of displacements, but reyes renderers are slow when it comes to GI...
REYES and GI have nothing to do with each other. Renderers with programmable shading can't take certain shortcuts during GI calculations unless the shader writer created the shader with GI in mind or the renderer wasn't to care about doing a sensible thing (thinking of physcal accuracy versus speed).
Since most RenderMan complinat renderers implement the programmable shading optional capability and most of them use REYES for camera "rays" (the term is extremely inadequate in this context), that leads people erratically believing REYES would be slow for GI.

It isn't, since REYES doesn't cover GI at all. How fast a renderer is when it comes to GI hence purely depends on the options & speed the GI/ray-tracing module of such a renderer offers and what use the TD (one would hope) using the renderer makes of these options and how they write their shaders -- provided the renderer offers programmable shading.
By the way the reyes renderers are renderman compliant renderers such as Aqsis, 3Delight, Air, and of coarse Prman...
AIR is not a REYES renderer and Aqsis doesn't support GI.


Cheers,

Moritz

[Mauritius]

I did check gelato out and they say it's quite fast for both motion blur and gi BUT you have to own a NVIDIA graphic card cause it's the first (that's what they say) engine using a 3d accelerator chip to render. My problem is that I've got a Radeon card.:sad:

Does vray get much slower when adding motion blur?

3Delight 4 was noticable faster than Gelato 1.2. I haven't looked at Gelato 2.0 vs 3Delight 5.0, but I would be surprised if 3Delight wasn't at least as fast -- mind that it doesn't use any GPU acceleration.

The first license of 3Delight (2 CPUs) is free and, unlike the "free" Gelato version, isn't crippled.
It also doesn't give a bloody **** what graphics card you have in your box or if you have graphics card at all. ;)

And like most REYES renderer, the speed hit with motion blur is around 25-50% vs no motion blur. Aka: a frame that renders 1 minute w/o will in the worst case render 1.5 minutes with motion blur. And we are talking true 3d motion blur here.

.mm

[Brian-Bradley]

Bit late on this but!!!

Mental Ray is capable of everything you have described here
and much more. Works great on network rendering, BUT! it
takes some doing to keep the render times down.

Vray has already been covered and I concur with my fellow
Max users about its capabilities.

Bri

[Nii]

Personally I am a Brazil Supporter =]. Mental Ray has always been a little too difficult for me to get the hang of.

[Ooze3d]

Ok so... If I want to save irradiance maps for fast GI then vRay, but my render times will duplicate or even triplicate with motion blur. I have the option of saving the velocity layer but that'll give me 2d motion blur and not as accurate as a normal 2d motion blur.

On the other hand, to get fast and real 3d motion blur I'll have to use a Renderman compilant renderer but then I'll have the GI problem...

I've also checked out Gelato renders and they don't look too photorealistic for me... Is that an user problem? I mean, can you get the kind of realism vRay does but rendering with Gelato? Cause I don't want to spend more money on an nvidia card to end up not using it.

Also I wanted to ask you about FinalRender and MentalRay. I don't care about moving objects anymore. I'll see what to do when I get to that point. But when it comes to flythroughs, do they perform better than vRay? Worse? Much worse? As I said, I'm willing to learn (speaking about MR).

I've got more questions but first I want to focus on a single renderer. By the way, thanks everybody for the answers. They do help a lot.

[Myliobatidae]

REYES and GI have nothing to do with each other. Renderers with programmable shading can't take certain shortcuts during GI calculations unless the shader writer created the shader with GI in mind or the renderer wasn't to care about doing a sensible thing (thinking of physcal accuracy versus speed).
Since most RenderMan complinat renderers implement the programmable shading optional capability and most of them use REYES for camera "rays" (the term is extremely inadequate in this context), that leads people erratically believing REYES would be slow for GI.

It isn't, since REYES doesn't cover GI at all. How fast a renderer is when it comes to GI hence purely depends on the options & speed the GI/rey-tracing module of such a renderer offers and what use the TD (one would hope) using the renderer makes of these options and how they write their shaders -- provided the renderer offers programmable shading.

AIR is not a REYES renderer.


Cheers,

Moritz

That's all very interesting, as I said "but reyes renderers are slow when it comes to GI..."

If that's incorrect, maybe you could point us toward one thats not slow at GI...

[Mauritius]

That's all very interesting, as I said "but reyes renderers are slow when it comes to GI..."
If that's incorrect, maybe you could point us toward one thats not slow at GI...
If you re-read my original post carefully it should become obvious to you that the answer is "any".

Give me an example scene + shaders that are slow to render in <insert you favorite GI + REYES capable renderer here>, and I'm sure I can point out to you what you need to change to make it faster or even "fast" (whatever that means).

I'm aware of the following renderers that use REYES and are GI capable:


3Delight
Pixie
Pixels 3D renderer
PRMan
VMantra
Given the broad statement you made originally, I guess you must have extensive experience that backs it up with at least three or four of those to justify it.

.mm

[ExP]

Mortiz,
you seem like you have a good background on the issue. Let me ask you this:
AIR is not a REYES renderer? I thought that every Renderman compliant renderer was somewhat based on the REYES scanline algorithms.
I'd like to hear your thought on this...


Ging back to the main topic of GI, here are my thoughts. Coming form the Film VFX industry I can say that real GI is rarely used in big FX shots.
Regardless of the renderer, using GI in animations would most of the time prove inpractical, with huge impacts on rendering times. On the other end if trying to compromise, it could easily lead to unacceptable quality (flickering artifacts).

The most common approach around this, would be to employ a hybrid use of Ambient Occlusion and LightMap illumination (Baked GI Maps). This could usually save a lot of rendering time, at the cost of the "more realistic light simulation" offered by full GI. Now if you're doing architectural lighting simulations, than GI is the way to go, however for most other uses Ambient Occlusion and baked GI Maps are a great compromise.


About render engines I'm for Mray and PRman (and other Renderman renderers). Their highly customizable structure is a must when dealing with big animation jobs.

- For 3dsMax and GI, finalRender offers a really good approach, quite fast render times, an hybrid GI/AmbientOcclusion engine, and 3D motion blur.

- I've heard good things about Brazil and GI, but its unacceptable motuion blur options (only 2D vector), made me discard it immediately.

- Vray is also nice, but it's way too "closed" (not really customizable) for my taste.

-Gelato 2 is shaping up to be a great Pre-Viz tool. However, to use it at its best you need an ultra high-end dual Quadro FX (in SLI mode) GPU rig.

Just my opinions...

Cheers.

[Mauritius]

Mortiz,
you seem like you have a good background on the issue. Let me ask you this:
AIR is not a REYES renderer? I thought that every Renderman compliant renderer was somewhat based on the REYES scanline algorithms.
I'd like to hear your thought on this...
AIR uses curvature base tesselation into triangles (vs. screen space based tessleation into quads that REYES uses). Unlike REYES, shading in AIR in independent of tesselation, shading samples are spaced evenly in screen space (in REYES they are 'as evely as it gets' based on surface tesselaton into &micro;-polys).
I haven't admittedly looked at AIR extensively for two years, but when I was testing it then, the approach had a lot of disadvantages. Particularly I remember that when going up with ShadingRate, your image would get blocky (like running a mosaic filter in Photoshop). We render with high shading rates all the time during lighting tests and the blurring effect this has on texture etc. in a REYES renderer is acceptable. Blocky image aren't -- particularly not if you have a VFX sup from the client on site looking over your shoulder. ;)

The other issue with AIR is that they don't try too hard to be 100% compatible with PRMan. Aka: if I write a shader that uses alot of new PRMan features, it won't compile with AIR and vice versa. 3Delight, Aqsis and Pixie try to stay as close to PRMan as possible. But then there's the Exluna incident and Sitexgraphics is located in Texas... Aqsis and Pixie are OSS and 3Delight's DNASoft is Montreal-based. So it's kinda understandable.
Ging back to the main topic of GI, here are my thoughts. Coming form the Film VFX industry I can say that real GI is rarely used in big FX shots. [...]
The most common approach around this, would be to employ a hybrid use of Ambient Occlusion and LightMap illumination (Baked GI Maps). This could usually save a lot of rendering time, at the cost of the "more realistic light simulation" offered by full GI. Now if you're doing architectural lighting simulations, than GI is the way to go, however for most other uses Ambient Occlusion and baked GI Maps are a great compromise.

That's what most places do these days. We currently bake ambient occlusion and bent normals into uv-based texture maps. We look into switching to brick maps soon to have a generalized pipeline that works with unparameterzied geometry (e.g. subdivision surfaces w/o any uv sets on them), as soon as 3Delight supports this feature.

About render engines I'm for Mray and PRman (and other Renderman renderers). Their highly customizable structure is a must when dealing with big animation jobs.
- For 3dsMax and GI, finalRender offers a really good approach, quite fast render times, an hybrid GI/AmbientOcclusion engine, and 3D motion blur.

- I've heard good things about Brazil and GI, but its unacceptable motuion blur options (only 2D vector), made me discard it immediately.

- Vray is also nice, but it's way too "closed" (not really customizable) for my taste.
All these renderers are either too "black-boxed" (lacking openess) or too immature at advanced features like programmable shading. The biggest advance when investing into a shitload of licenses of a RenderMan compliant renderer is that you can switch to a competitor's product (or mix with it) almost instantly in your pipeline.
-Gelato 2 is shaping up to be a great Pre-Viz tool. However, to use it at its best you need an ultra high-end dual Quadro FX (in SLI mode) GPU rig.
And if you use Gelato, perpare to spend a "bit" of money on keeping your farm cool.

When I compared it with 3Delight last March, 3Delight kicked its ass (w/o any hardware acceleration, mind you). It also had quite afew of the issues that Entropy had...
But I'm keen to play with v2. Anyway, as long as they don't remove the hardware dependecy, I don't think many places will invest a huge amount of money into that renderer. I'd also think that it is mereley interesting for nVidia from a marketing pov (as in "our hardware is used to drive blockbuster vfx"). It's a shame what happened to Exluna; Entropy could be a solid PRMan competitor by now but from personal my pov, Gelato currently isn't.


Cheers,

Moritz

[ExP]

Mortiz,

Thanks for your detailed explanation on the Air renderer. I have never had a chance to test it in depth, but your comments seem to have a solid base.


It's a shame what happened to Exluna; Entropy could be a solid PRMan competitor by now but from personal my pov, Gelato currently isn't.


-I definitely agree on this. And considering the great job the guys from Exluna did on Entopy, I am hoping they can bring some good things to Gelato in the future. On the other end, I also hope this doesn't end up being just another marketing tool for nVidia...

-It seems like you guys at Raising Sun Pictures are putting some serious weight on 3Delight. I will definitely give it more consideration from now on. I just hope I could get my hands on a great tool like Affogato, as we are currently switching to XSI (hopefully the team at Graphic Primitives will pull something good our of XSIMan).

-Are you seeing any major disadvantages using 3Dlight rather than PRman? The price tag is definitely a Pro for 3Delight, what are the main Cons?

Thanks!

P.S. sorry everybody, if I'm getting our of topic here...

[Ooze3d]

P.S. sorry everybody, if I'm getting our of topic here...

It's not out of topic really... We're still talking about animation and production quality render. I started this thread so I could choose a good renderer for my needs. Of course I didn't expect to get an easy answer cause there's not a 'perfect' renderer for every situation and everyone has his/her own preferences but I've learnt a lot and I think now I'm closer to the render engine I need.

So go on!

[Mauritius]

-It seems like you guys at Raising Sun Pictures are putting some serious weight on 3Delight. I will definitely give it more consideration from now on.
There are many places that know what a good alternative 3Delight is to PRMan.
If you check out the Wikipedia entry (http://en.wikipedia.org/wiki/3Delight) for the renderer, there's a list of movies it was used on.

In high end VFX, no one makes a fuss about what renderer they use. PRMan has been the renderer of choice for 90% of the blockbuster VFX (and the majority of full CG animated features) since it became available. Yet may people still believe the T-Rex in Jurassic park had so many million polygons (it was, in fact, all NURB patches) and was rendered with some super secret ILM internal render engine.
If you want to invest in a render farm with a couple of hundred licenses, a lot of factors come into play. "How well does a rendere handle GI?" is usually of least concern.

I've been using 3Delight in the past whenever I worked at a place that had a limited number of PRMan seats and that number didn't suffice to turn a shot around in the time given. Now that the renderer is no longer free, that is not possible anymore. However, during these years I found that DNASoft has the best support I ever saw from a 3D software vendor.
I do imagine that Sitexgraophics would be pretty close as they are a small company too and AIR's author, Scott, was very quick with fixes & features when I was beta testing the renderer a few years back.

Our turnaround times for bugfixes from 3Delight are usually less than 12 hours. Most of the time this is rather glitches than bugs (meaning you could work around them easily). We had two bugs with 'stable features' where the renderer would crash in rare situations during the last year of production here at RSP Adelaide and they both got fixed within a day.
On the feature side, DNASoft have implemented over a dozen features for this particuar production that make using 3Delight more fun and solve issues for which we would have otherwise had to develop a custom DSO in house (aka spend dev time, plus other users of the renderer wouldn't have access to it afterwards).
All other hard crash bugs we had were with new features we requested and also got fixed in no time.
Summing up I can say that using 3Delight has been a great experience.

People also commonly think that fast GI support and being able to handle milliopns of polygons are features of paramount importance (which they are if you use a non-REYES renderer for monkey-pushbutton-instant-GI-interior-scenes(tm) that require minor to zero knowledge about rendering), whereas the truth is that robustness, openness, and being able to handle millions of high level primitives (like true NURB patches or true subdivision surfaces) are of much higher importance that GI & polygons are and ever will be in the forseeable future.
I just hope I could get my hands on a great tool like Affogato, as we are currently switching to XSI (hopefully the team at Graphic Primitives will pull something good our of XSIMan).
Affogato is a lot like Liquid, so unless you build a pipeline around it, it is not much use for an end user.
I personally would also strongly counsel against using XSI at the rendering stage of a feature VFX pipeline if you use a RMan compliant renderer. XSI's strength is animation, that's where it belongs. Whether you have XSIMan, Affogato or another plug-in like that doesn't make much difference.

-Are you seeing any major disadvantages using 3Dlight rather than PRman? The price tag is definitely a Pro for 3Delight, what are the main Cons?
DNASoft has not remotely as many customers as has Pixar. Hence it is easier to get to talk to the developers and get what you want. The other thing is the rapid release cycle. With the stability of the renderer and the speed that features we request get implemented, rather than locking ourself into a version of the renderer for the length of a production, we adapted to this rapid release cycle. So we usually have a 3Delight version used on the floor and another or several newer ones that are a few weeks ahead of that ahich you can use by explicitly requesting them when rendering. As soon as we are sure that a newer version has only fixed and added things and not broken any working ones, we make that version the current one. So far that has worked great.

3Delight also put some features in for us that aren't in PRMan at all and that would have made our pipeline a lot more complex on the last production. Particularly these are exclusive AOVs with explicit control over matting. I heard something like that will be in the next PRMan too, but we asked for and got this feature last year when we needed it. Pixar takes ages to put feature requests in (at least they used to).

PRMan still is almost a decade older than 3Delight. Aka: you don't make a mistake investing into that renderer ever.
So if you are rendering/CG supervsior and you have to make a decision but never used 3Delight, chances are you'll pick PRMan for a plain lack of facts about and experience with the former renderer.
It's kinda like in business. You know the famous quote "No one ever got fired for suggesting to buy IBM [hardware/software/consulting sevices]". :)

Cheers,

Moritz

[ExP]

Yet may people still believe the T-Rex in Jurassic park had so many million polygons (it was, in fact, all NURB patches) and was rendered with some super secret ILM internal render engine.


-Hahah... :thumbsup:!

"How well does a rendere handle GI?" is usually of least concern.

I totally second that. I have yet to work on a show that actually uses full GI in production.


I personally would also strongly counsel against using XSI at the rendering stage of a feature VFX pipeline if you use a RMan compliant renderer.


In our studio we are currently considering embracing XSI for animation and are evaluating the inclusion of Renderman in our renderfarm. I have to say that from the tests we've been conducting so far, it has definitely been a little "rough." Especially the implementation form XSI to Renderman.
We are just coming off a project that required huge crowd animations, most of it rendered in MRay (you can imagine the memory allocation headhaches). This made us finally decide to re-evaluate our pipeline. I'm personally just plunging into the Renderman world, but I can say that just the way it handles Motion Blur and SubSurfices is worth the switch.

Now I feel really bad we have not implemented RenderMan before! However our studio started on a 3dsMax/Maya/MRay workflow and never really had the time and resources to expand/refine our pipeline. Especially when including RenderMan, which let's jsust say requires some "special attention."


So if you are rendering/CG supervsior and you have to make a decision but never used 3Delight, chances are you'll pick PRMan for a plain lack of facts about and experience with the former renderer.
It's kinda like in business. You know the famous quote "No one ever got fired for suggesting to buy IBM [hardware/software/consulting sevices]".


Hahaha... That is so true! Unfortunately people do choose a certain tool over the other just because of its "name." That's why in our case we are trying to evaluate all of our options before making the final decision.
I personally think that the more intimate/faster support we would get with a smaller company like DNASoft is definitely a big plus!

Thanks for all of your great advice.

[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.