PDA

View Full Version : Creeepy...



SonicN2O
09-09-2010, 09:55 AM
I looked it up, and the SFX for Avatar was done in Lightwave (yay!) 3ds max and maya (:lightwave) Vue (not suprising) and...

Blender????:eek:

I used blender before I used lightwave. one describing word:
CRAPPY, CRAPPY, CRAPPY.

Dreamcube017
09-09-2010, 11:34 AM
Wow really??? That's new. It's like they used every 3D app under the sun.

And come on, Lightwave's betterthan blender, but blender's not THAT bad.

oliversimonnet
09-09-2010, 11:42 AM
haha wehad a thread on this a while back hahah
and dude give the other 3D aps some credit :)

SonicN2O
09-09-2010, 11:46 AM
Wow really??? That's new. It's like they used every 3D app under the sun.

And come on, Lightwave's betterthan blender, but blender's not THAT bad.

If you have LightWave, Vue Xstream, After Effects, and Final Cut Pro, There's NO OTHER SOFTWARE you need for movie making; even for something like Avatar.

Dreamcube017
09-09-2010, 12:13 PM
Hm I suppose. I personally don't like Final Cut Pro, but yeah, Avatar pretty much SCREAMS Vue andI can see lightwave too... but I don't like Ater effects that much either... mainly because of how it runs on most computers. (slow)

oobievision
09-09-2010, 12:15 PM
well buying Flame and Nuke is usually out of the question

bazsa73
09-09-2010, 12:22 PM
If you have LightWave, Vue Xstream, After Effects, and Final Cut Pro, There's NO OTHER SOFTWARE you need for movie making; even for something like Avatar.

Why did those guys then working on Avatar used other softwares, ones that are not included in this list?

RebelHill
09-09-2010, 12:25 PM
If you have LightWave, Vue Xstream, After Effects, and Final Cut Pro, There's NO OTHER SOFTWARE you need for movie making; even for something like Avatar.

I think you'll find thats a little bit of a stretch.

Actually no, its just plain wrong, and ill informed.

Theres a lot of off the shelf software used in big movies of taht nature, but the vast majority of what you see on screen is based on propietary solutions that are developed in house by the companies making these movies.

OnlineRender
09-09-2010, 12:34 PM
Hm I suppose. I personally don't like Final Cut Pro, but yeah, Avatar pretty much SCREAMS Vue andI can see lightwave too... but I don't like Ater effects that much either... mainly because of how it runs on most computers. (slow)

Say that too Rob and other artists that worked on the production , "it SCREAMS VUE " . chuckles , I also don't think they rendered out in Vue ,or used Lightwave in the final CUT, I also disagree on the After Effects comment ,I run AE on 4gig and it runs cool .

DragonFist
09-09-2010, 03:40 PM
I forget where, but in a video on the virtual camera stuff, it was gone over how most of the content was done in Lightwave but other packages were used as well. Blender was specifically mentioned as being used for a particular tree that one of Blender's tools made it easy to make in Blender where it was harder to do in other packages. I got the idea that that was pretty much its involvement.

SonicN2O
09-09-2010, 06:14 PM
Hmmm... you could be right. btw, how did the "wtf" thread get closed?

SonicN2O
09-09-2010, 06:14 PM
also, is there a way to import a vue xstream file into lightwave?

OnlineRender
09-10-2010, 01:12 AM
also, is there a way to import a vue xstream file into lightwave?

In Vue 8xstream , you go Export Scene - LWO , but I found it messed the scale and lost the textures , you can also export as 3Ds but I also found it centred all objects and puts them into the 1 layer .

I did find a nice work around that goes LW -Blender- MAX - Back to Blender - VUE .

oliversimonnet
09-10-2010, 05:03 AM
Hmmm... you could be right. btw, how did the "wtf" thread get closed?
it got closed by NT(i think)

NanoGator
09-10-2010, 03:23 PM
I've been told that Vue was occasionally used with 'distant jungle' shots. But inside the jungle itself, no, those trees/plants/moss etc were all placed by hand.

jrandom
09-10-2010, 04:15 PM
I think my biggest question about Avatar is what renderer did they use, what type of illumination (MC, FG, mixed, etc...) and how much computing power did they have to throw at it?

I ask because as I dive into doing animations with GI I've discovered that the settings one has to use to eliminate flicker generally result in ridiculously long render times (y'know, for scenes with so much movement that you can't pre-bake the radiosity).

shrox
09-10-2010, 04:30 PM
I think my biggest question about Avatar is what renderer did they use, what type of illumination (MC, FG, mixed, etc...) and how much computing power did they have to throw at it?

I ask because as I dive into doing animations with GI I've discovered that the settings one has to use to eliminate flicker generally result in ridiculously long render times (y'know, for scenes with so much movement that you can't pre-bake the radiosity).

Yeppers, my scenes often crash on just the radiosity cache.

jrandom
09-10-2010, 04:33 PM
Oh, I see. A lot. (http://www.datacenterknowledge.com/archives/2009/12/22/the-data-crunching-powerhouse-behind-avatar/)

Intuition
09-10-2010, 04:42 PM
I think my biggest question about Avatar is what renderer did they use, what type of illumination (MC, FG, mixed, etc...) and how much computing power did they have to throw at it?

I ask because as I dive into doing animations with GI I've discovered that the settings one has to use to eliminate flicker generally result in ridiculously long render times (y'know, for scenes with so much movement that you can't pre-bake the radiosity).

I did a few shots in maya for the coke zero ads for Avatar. I recieved assets from weta that were in maya format that had rib shaders attached to the surfaces and fur which meant Renderman was being used. Even the hypershade references had renderman names.

jrandom
09-10-2010, 04:47 PM
I'm wondering if I should just buy "Physically Based Rendering" (http://www.amazon.com/Physically-Based-Rendering-Implementation-Interactive/dp/012553180X) and try to write a rendering plugin that is suited for animation.

I'll probably fail, but it certainly would teach me a ton about how these renderers work and how to approach the settings they have.

(Goal #2 would be to try to make a Vue-like plugin. E-On software has this horrible tendency to try to squeeze as much money as they can from their userbase for minor upgrades and crashy software. I've never seen so many people hate a company that makes the software they use, and I can only attribute this to the fact that Vue doesn't have any serious competition yet.)

NanoGator
09-10-2010, 04:48 PM
I've worked on a bunch of shows requiring photo-real elements (often buildings) to be put into a shot. Rarely are radiosity renders of the animation done. Instead we implement clever uses of projection mapping. On a number of occasions I've rendered a simple model with appropriate lighting and handed it off to matte painters to use that as a starting point then take it to 'real'. Those images are re-projected back onto geometry and tracked back into the scene.

I'm pretty certain a number of floating mountain shots were done that way on Avatar. It's a practical and efficient way to work, it doesn't matter which app you use at that point.

NanoGator
09-10-2010, 05:50 PM
Dang, I really assassinated this thread. O_o

Mr Rid
09-11-2010, 04:37 AM
I'm wondering if I should just buy "Physically Based Rendering" (http://www.amazon.com/Physically-Based-Rendering-Implementation-Interactive/dp/012553180X) and try to write a rendering plugin that is suited for animation....

Just write a conversion plug for Lightwave-to-LuxRender. Sell it and make some dough re mi.

silviotoledo
09-11-2010, 07:52 AM
Rendering Power? Only 40.000 CPUS ( around 5k 8 cores i7 processors ) at Weta Digital. Other companies, like ILM have another similar amount of machines to render.

No, lightwave can't do that only because Lightwave doesn't have:
Layer Animation - Mocap retargeting - That Cool Dinamics and Softbodies - Muscles

Rendering is probably Mental Ray ( and maybe some renderman ) with a lot of compositing.

Captain Obvious
09-11-2010, 08:00 AM
If you have LightWave, Vue Xstream, After Effects, and Final Cut Pro, There's NO OTHER SOFTWARE you need for movie making; even for something like Avatar.
You have got to be kidding me. That's like saying that as long as you have a forest filled with trees, a handsaw, some nails and a hammer you can build a skyscraper.



Edit: And Blender is actually really good, if you manage to get your head around the atrociously weird user interface...

jrandom
09-11-2010, 09:36 AM
Just write a conversion plug for Lightwave-to-LuxRender. Sell it and make some dough re mi.

Lux is absolutely beautiful, but what I'm looking for is a renderer that is fast and good for flicker/shimmer-free animations, even at the sacrifice of physical correctness.

Rather than complain too much about existing renderers I figure why not try to write my own and see how hard it really is. There's got to be some solution for us peons that don't have huge render farms at our disposal.

SonicN2O
09-11-2010, 10:26 AM
Just write a conversion plug for Lightwave-to-LuxRender. Sell it and make some dough re mi.

OK, just give me a free copy. :hey::boogiedow

Dreamcube017
09-11-2010, 02:33 PM
A free copy of the plugin that someone will write, or a free copy of LuxRender... Lux is free... it's not even at version 1 yet.

JeffrySG
09-11-2010, 03:52 PM
I think you'll find thats a little bit of a stretch.

Actually no, its just plain wrong, and ill informed.

Theres a lot of off the shelf software used in big movies of taht nature, but the vast majority of what you see on screen is based on propietary solutions that are developed in house by the companies making these movies.

:agree:
Also, if you do some research you'll see that Blender is actually a very capable piece of software. There are a few threads already talking about how many great features it has. But people do some very nice/amazing work in Blender. This thread just seems like a way to troll a bit.

Dreamcube017
09-11-2010, 05:24 PM
I have to agree... the comment on blender seemed somewhat fanboyish even it it wasn't meant that way.

Blender 2.5's UI is a bit better... kinda like lightwave/XSI a bit.

And plus blender has a great free book to help you around the old interface. Depending on what I'm doing, I still use blender to do some modeling.

I like lightwave's set of modeling toosl though too, and I like it's material and lighting set ups.

I never swear to one single piece of 3D softwware. I still take a look at the new Maya, 3DS, blender, Houdini, whatever else I want. But I still love Lightwave, even though I'm pretty new to it.

Oh and yeah, they use a lot of proprietary software to do stuff... now that being said, with some work and time (and money) you CAN make some very VERY good things VERY close to avatar quality. It'll just cost you money, time, and lots of work... or fun. Depends on which way you see it.

Heck, half the work's done for you sometimes if you just want to get some very realistic characters without having to model them. DAZ does have some REALLY nice models for some great prices... and they have an FBX exporter so you CAN bring them into lightwave (and Vue and whatever else) so there's nothing really stopping you.

Mr Rid
09-11-2010, 06:37 PM
Lux is absolutely beautiful, but what I'm looking for is a renderer that is fast and good for flicker/shimmer-free animations, even at the sacrifice of physical correctness....

I actually dont have trouble getting economically interpolated and flickerless GI animations when using motion blur in LW. Each of at least nine or more interpolated passes can smerge nicely into a smooth render. And I find that the Classic Camera nearly always yields faster render times with equal or better AA quality than the perspective camera that most presume to use.

There is also a common misconception that an animated, baked radiosity will save render time. But I have only ever seen it take longer when compared to the same settings unbaked.

Captain Obvious
09-12-2010, 02:15 AM
And I find that the Classic Camera nearly always yields faster render times with equal or better AA quality than the perspective camera that most presume to use.
That really depends on what you're rendering, though. The type of scenes I usually end up with render much more quickly with the perspective camera.

Mr Rid
09-12-2010, 03:39 AM
That really depends on what you're rendering, though. The type of scenes I usually end up with render much more quickly with the perspective camera.

I'd like to see a practical example of one of these scenes some day. I keep hearing the P cam is faster for something but I have yet to run into it.

Captain Obvious
09-12-2010, 05:19 AM
I'd like to see a practical example of one of these scenes some day. I keep hearing the P cam is faster for something but I have yet to run into it.
Basically, have a look at Cityscape's website and pick any render done in Lightwave and I can pretty much darn well guarantee that the perspective camera was faster. :)

The stuff I did at Digiguys was generally way faster with the perspective camera, too, because of the heavily ray traced shading.

MrWyatt
09-12-2010, 01:10 PM
Rendering Power? Only 40.000 CPUS ( around 5k 8 cores i7 processors ) at Weta Digital. Other companies, like ILM have another similar amount of machines to render.

No, lightwave can't do that only because Lightwave doesn't have:
Layer Animation - Mocap retargeting - That Cool Dinamics and Softbodies - Muscles

Rendering is probably Mental Ray ( and maybe some renderman ) with a lot of compositing.

Rendering is mostly renderman with a mix of proprietary gpu prepass rendering that was used to bake out the indirect illumination. but anything that gets rendered at weta is pretty much all rendered using prman from pixar.

Mr Rid
09-12-2010, 07:00 PM
Basically, have a look at Cityscape's website and pick any render done in Lightwave and I can pretty much darn well guarantee that the perspective camera was faster. :)

The stuff I did at Digiguys was generally way faster with the perspective camera, too, because of the heavily ray traced shading.

Renders dont demonstrate what is optimized by P cam. I would like to see an actual scene. Several times now, I noticed that someone I work with has a scene that seems to be taking too long to render on the stack. I load the scene and the P cam is being used because the artist believes it is rendering faster. In most cases, I come up with Classic settings that render much faster with equal result, and in fewer cases I find settings that render in close to the same time. But I have never found a scene that rendered significantly faster with the P cam. When I need many passes of blur, I find that the P cam is always much slower to get a same or better result than with C cam. I find that the artist just doesnt quite understand how AA, blur and recursions work in both cameras and how to get an equal result.

I keep hearing that heavy tracing, bounces, and multi-threading is somehow optimized in the P cam but that has never been the case in any of my scenes, or in scenes posted on this forum. So I keep wondering what specific render situation makes a practical difference.

Captain Obvious
09-12-2010, 07:58 PM
Likewise. I'd like to see a scene that renders faster in the classic camera. We obviously render very different things! :)

jrandom
09-12-2010, 10:16 PM
I threw together a little scene to put this to the test. The .zip file also contains all the test renders as well. (There's a five-file limit for attachments on posts, so I couldn't put them all here in the message.)

This was done on a 12-core/24-thread Mac Pro. "Classic" reconstruction was used for both cameras.


Regular AA:

Classic and Perspective cameras deliver the same quality, with Perspective coming in at slightly faster (1m 15s compared to Classic's 1m 25s). If you're rending an animation this could shave a considerable amount off the total render time.

Classic's AA was set to PLD-3, and Perspective's AA was also set to "3".


Adaptive Sampling AA:

At the same settings (but with an AS threshold of 0.025), Classic at first appears to beat the pants off of Perspective, clocking in at 1m 3s compared to Perspective's 2m 19s, but the quality of the Perspective Camera's output is clearly superior (look in the 'Render' folder in the .zip for this image). I reduced its AA to 1 and played around with the AS threshold to get it as close as I could to Classic's render and Perspective's time on those settings come in at 55s.


As far as I can tell, if you've got blurry reflections in your image and are rendering on a multi-core system then the Perspective camera is definitely the way to go.

Cageman
09-13-2010, 05:20 PM
Multithreading IS way better in Perspective camera.

Lets say you have a scene with lots of reflections in the lower half of the image, only black in the upper part of the image. Classic camera will not repurpose a thread that rendered the much faster upper part of the image, while Perspective and Real Lens cameras will divide whatever is left on any thread and let previously finished renderthreads to help out.

What you can do though, to minimize the negative effect of using Classic Camera in such a situation is to use more threads than you have with your machine, but it is not as effective as actually using Perspective Camera or Real Lens camera.

Cageman
09-13-2010, 05:26 PM
I keep hearing that heavy tracing, bounces, and multi-threading is somehow optimized in the P cam but that has never been the case in any of my scenes, or in scenes posted on this forum. So I keep wondering what specific render situation makes a practical difference.

Actually, someone posted a Perspective Cam optimized version of the first Creature kit, which renders way faster than Classic Camera but with very similar results.

Mr Rid
09-15-2010, 12:25 AM
Am rendering as 16 threads on an i7 960, 3.2GHz, 12gb ram.

Here's a new version of jrandom's CamRoom scene with 3 cameras for comparison, rendering only the right side of frame. 88105
First camera is Perspective and the second is Classic. Both render with about equal AA results, but the Perspective Cam took 104 seconds for me, and the Classic took 86 seconds.
gif compare- 88108

The third camera is same as the second Classic camera but with Sampling at .03 instead of .04. The render time was 107 seconds but has markedly less noise.
gif compare- 88107

So again I see that the Classic cam renders a little faster with the same AA, or can give better AA at about the same render time. I dont see that multi-threading makes any difference. Then if I turn on 5 blur passes for both camera types to squelch messy interploated GI, the Perspective camera takes three times longer than Classic.

With the Vampires Suck baby, the motion was so rapid in places that I needed many blur passes to kill steppys. At 8 passes of photoreal blur with the Perspective Camera there were very visible steppys, but in the same render time I could render 66 passes with Enhanced X, dithered Classic and get much better result.

Here is another example scene with a buncha reflections and rays bouncing all over that are supposedly optimized by the Perspective camera.88106

Again, I get the same render time with either Perspective or Classic but with better AA quality in the Classic. There is more noise and jaggies in the Perspective Camera. Anything I do to get the Perspective Camera AA to look as good as the Classic will result in longer render times.
88110

So what am I missing? I dont see the Perspective Camera render faster than Classic Camera with equal AA result, regardless of lotsa rays or multithreading. I also find that the Classic Cam's Enhanced settings work better than any other AA settings I can come up with in LW. However I have seen that Classic AA has more difficulty with certain reflection noise.

jrandom
09-15-2010, 12:29 AM
How many hardware threads does your machine support? Maybe the perspective camera advantage only really kicks in under those conditions?

Mr Rid
09-15-2010, 12:48 AM
How many hardware threads does your machine support? Maybe the perspective camera advantage only really kicks in under those conditions?
8. Although renders only speed up when I set LW to 16 for whatever reason. I also never saw any advantage in Perspective camera renders on the MacPro quads at work either. Do you get a similar time comparison with the reflection scene?

88112

JonW
09-15-2010, 01:10 AM
LW 10 supports a lot more threads. With a pair of X5600 CPUs, 8 of the 24 threads will be idle at any one time.

toby
09-15-2010, 03:08 AM
So what am I missing? I dont see the Perspective Camera render faster than Classic Camera with equal AA result, regardless of lotsa rays or multithreading. I also find that the Classic Cam's Enhanced settings work better than any other AA settings I can come up with in LW. However I have seen that Classic AA has more difficulty with certain reflection noise.
This could be the same thing that was true with "Raytrace Optimization", the feature that disappeared at the same time the perspective cam was introduced.

At Digital Domain commercials at the time, the sentiment was "it's a complete waste of time", but I found out when it was worthwhile; for scenes that were so long that no one would bother testing it for. In that case we were rendering whole cities of glass. One pass of aa would take 1 1/2 hrs with dd's default settings; no raytrace optimization, but only 1 hour with it on. It was the overhead involved in the optimization that made it always seem slower, because in 1 minute tests it definitely was. But if an overhead of 2 minutes saves you 30% on a 4 hour frame, hell yea it's faster. I also remember that if we saved the scene in the then new lw9 and opened it in lw8, raytrace optimization would always be back 'on'.

Another factor; as I recall the 'enhanced' aa is equal to 0.7 oversample in the perspective cam, which in this scene speeds up it's render by 10-15%, by virtue of being softer/less jaggies for the adaptive to detect. But, jrandom, your test seems to show more graininess for the faster persp cam renders.

to be continued (god I love testing renderers!)

Lightwolf
09-15-2010, 03:55 AM
With the Vampires Suck baby, the motion was so rapid in places that I needed many blur passes to kill steppys. At 8 passes of photoreal blur with the Perspective Camera there were very visible steppys, but in the same render time I could render 66 passes with Enhanced X, dithered Classic and get much better result.
Why did you use multiple passes with photoreal MBlur, as opposed to 1 pass, a low base AA and mid-range AS threshold?

Cheers,
Mike

Lightwolf
09-15-2010, 03:57 AM
This could be the same thing that was true with "Raytrace Optimization", the feature that disappeared at the same time the perspective cam was introduced.
And all it basically did was make tighter bounding boxes around the geometry (which required scanning all vertices per frame, thus the pre-process).

Cheers,
Mike

lwaddict
09-15-2010, 07:08 AM
If you have LightWave, Vue Xstream, After Effects, and Final Cut Pro, There's NO OTHER SOFTWARE you need for movie making; even for something like Avatar.

Uh...
well...
not exactly.

You should get your feet wet before discussing how great the water is. 8/

Not slamming ya, we've all been there.

yaschan
09-15-2010, 07:32 AM
8. Although renders only speed up when I set LW to 16 for whatever reason. I also never saw any advantage in Perspective camera renders on the MacPro quads at work either. Do you get a similar time comparison with the reflection scene?
88112

Mac Pro Quad here too. And I wonder exactly the same.

(edited: after I read the rest of the posts..)

Captain Obvious
09-15-2010, 08:28 AM
With the Vampires Suck baby, the motion was so rapid in places that I needed many blur passes to kill steppys.

Classic cam, dithered motion blur, 66 passes in total. About 7 seconds.
http://www.newtek.com/forums/attachment.php?attachmentid=88115&stc=1&d=1284560740

Perspective cam, classic motion blur, 66 passes in total. About 7 seconds.
http://www.newtek.com/forums/attachment.php?attachmentid=88116&stc=1&d=1284560697

Perspective cam, photoreal motion blur, a combination of low base AA, adaptive sampling and a few passes. About 5 seconds. I'd say this one is way better.
http://www.newtek.com/forums/attachment.php?attachmentid=88117&stc=1&d=1284560734



The scene is a human character, parented to a null. The null moves diagonally across the screen in the space of one shutter opening, and rotates 360 degrees around all axes.

toby
09-15-2010, 09:56 PM
Classic cam, dithered motion blur, 66 passes in total. About 7 seconds.
-
Perspective cam, classic motion blur, 66 passes in total. About 7 seconds.
-
Perspective cam, photoreal motion blur, a combination of low base AA, adaptive sampling and a few passes. About 5 seconds. I'd say this one is way better.
-



The scene is a human character, parented to a null. The null moves diagonally across the screen in the space of one shutter opening, and rotates 360 degrees around all axes.
Score one for the persp cam. What are the final settings exactly?

One thing I prefer about the persp cam is the flexibility. You can mimic the classic cam or do things it can't. Any number of passes is possible, you're never stuck with either 17 or 33 passes, which can save you rendertime. You can go waaay beyond 66 passes if you need a certain effect.

You can also do little limited region renders and all 8 cores will still be used, whereas with the classic cam you're stuck with the procs that are assigned to that area and no more. There's enough things like that to make it better overall imho.

Mr Rid
09-15-2010, 10:45 PM
Why did you use multiple passes with photoreal MBlur, as opposed to 1 pass, a low base AA and mid-range AS threshold?

1 pass would not smooth out interpolated FG flicker. Another odd thing I forgot about those baby scenes was that 1 pass of photoreal blur took almost twice as long as 2 or several passes. And several passes would render in the same amount of time as only a few passes. I didnt have time to troubleshoot it, but I think the SSS preprocessing/processing was mysteriously confusing things, since that was the one unusual thing I had never used before, and I had other odd problems with.

btw, can you give an example of AA, AS values you refer to?


...What are the final settings exactly?
I'd like to know too.

toby
09-15-2010, 11:25 PM
Ok, for a pure raytrace comparison I grabbed the old Raytrace benchmark scene and cranked up the trace settings to match today's machines' power; 64 bounces, 2k, and reflection/refraction blurring. No aa for either cam (that's another test entirely)

Classic - 16 threads - 3 min, 48 sec

Persp - 8 threads - 3 min, 10 sec

about 20% faster, pretty good

jrandom
09-15-2010, 11:29 PM
Where is this benchmark scene?

toby
09-15-2010, 11:45 PM
Where is this benchmark scene?
In the LW content directory that comes with the software. Set your content directory to Content, the navigate to Scene/Benchmark.

Captain Obvious
09-16-2010, 03:19 AM
What are the final settings exactly?
I didn't save the scene so I can't check, but I think it was 5 base AA, 0.02 adaptive threshold, 3 motion blur passes.


Combining motion blur passes with adaptive sampling is actually a really useful method, even when you're not using motion blur. It often gives better results than higher base AA.

lardbros
09-16-2010, 03:36 AM
Interesting thread... since it's been hijacked!

I own Vue, Lightwave and a few other bits of the software mentioned, and I certainly couldn't create avatar :) In a million years!

jrandom
09-16-2010, 09:00 AM
Interesting thread... since it's been hijacked!

I own Vue, Lightwave and a few other bits of the software mentioned, and I certainly couldn't create avatar :) In a million years!

Actually, you could. And it would take you a million years. (Note: That's just rendering time.)

Captain Obvious
09-16-2010, 09:26 AM
Actually, you could. And it would take you a million years. (Note: That's just rendering time.)
Yeah, that's the thing.

At the peak of activity, WETA had about 400 artists working on Avatar. Now, if that peak lasted for... say... a year, you could then imagine that it would take a single person at least 400 years to get it done, and that's not even accounting for the fact that nobody is as good at everything as a specialist is at their specialty.

Lightwolf
09-16-2010, 09:32 AM
btw, can you give an example of AA, AS values you refer to?

I usually render at AA 2-4 (that depends), AS 0.03, photoreal MBlur, Classic Filter, Fixed pattern. Oversampling 0.0125 if I use it at all.
That's for animations (mostly rigid body and technical).

Cheers,
Mike

Captain Obvious
09-16-2010, 09:54 AM
I usually render at AA 2-4 (that depends), AS 0.03, photoreal MBlur, Classic Filter, Fixed pattern. Oversampling 0.0125 if I use it at all.
That's for animations (mostly rigid body and technical).
For our print-res stills, it's often faster/better to use less adaptive sampling and more base AA. I usually render with... say... base 10 / adaptive 0.065 / fixed pattern / oversampling 0.5.

The oversampling softens the render, which is usually a good idea when you're rendering to a 20 megapixel photographic plate.

prometheus
09-16-2010, 02:07 PM
I'd like to see a practical example of one of these scenes some day. I keep hearing the P cam is faster for something but I have yet to run into it.

yay..I got my new machine up and running now, and finally I have a chance to test turbulence 4d, Ive noticed that a simple test render of 480x360
for a sooty fire render took 8.3 seconds with perspective cam antialiasing 1.
and with classic cam 20.4 seconds..this with 8 render threads in Automatic mode.

I switched to 16 render threads..classic was giving 10.9 seconds and here
the perspective was the same 8,3 seconds...just a little bit faster.

Michael

Cageman
09-16-2010, 06:17 PM
The speedoptimizations in Perspective Camera also scales with complexity of scene. Lots of heavy raytracing, lots of textures, lots of polygons will most likely give much better performance compared to classic camera, if setup properly.

Mr Rid
09-16-2010, 07:09 PM
...for a sooty fire render took 8.3 seconds with perspective cam antialiasing 1.
and with classic cam 20.4 seconds..this with 8 render threads in Automatic mode.

I switched to 16 render threads..classic was giving 10.9 seconds and here
the perspective was the same 8,3 seconds...just a little bit faster.

But how do the AA settings compare? Posting a scene is the best way to demonstrate a comparison.


Score one for the persp cam.

The only time I use Perspective is for extremely fast movement as in C.O.'s example where Classic passes wont accommodate (which is very rare in practicality), or for DOF blur (Classic is unusably poor, but DOF is usually better handled in post with a depth pass), or to quell some unusually intense noise that occurs in certain situations where Classic AA has more difficulty for some reason.


One thing I prefer about the persp cam is the flexibility. You can mimic the classic cam or do things it can't. Any number of passes is possible, you're never stuck with either 17 or 33 passes, which can save you rendertime. You can go waaay beyond 66 passes if you need a certain effect.

Actually the Classic camera max is 70 passes, but I dont know why use classic blur with Perspective if it renders with lesser quality AA in the same time as Classic, as in the posted examples.


You can also do little limited region renders and all 8 cores will still be used, whereas with the classic cam you're stuck with the procs that are assigned to that area and no more.

Limited region optimization that Cageman mentioned is a new one to me. But the two example scenes I posted are both using limited region and I see no advantage to the Perspective camera.


Ok, for a pure raytrace comparison I grabbed the old Raytrace benchmark scene and cranked up the trace settings to match today's machines' power; 64 bounces, 2k, and reflection/refraction blurring. No aa for either cam (that's another test entirely) .

I cant think of a practical situation rendering ray traced objects and no AA so I dont know what this demonstrates.

My point here is trying to figure out why so many users maintain that the Perspective Camera is optimized to render faster than the Classic in some regard to reflections, recursions, multi-threading, and now limited region. So far, I have only seen scenes that were not faster as a coworker or forum-goer at first believed. What the user perceived as 'faster' was actually with lesser quality AA settings that the user was not comparing closely enough. Or there is some particular situation that no one has yet to produce as an example.

AlI I read in the LW docs about this is "More complex scenes tend to render faster with the Perspective Camera." But I cant find what constitutes a scene 'complex' enough to render a practical difference.

Cageman
09-16-2010, 07:26 PM
Limited region optimization that Cageman mentioned is a new one to me. But the two example scenes I posted are both using limited region and I see no advantage to the Perspective camera.

I never said that...

Perspective Camera will repurpose a finished thread to continue help with rendering while Classic Camera does not. This is why classic Camera ends up being alot faster if you use more threads than what your CPU really has. It clearly explains why Prometheus experienced 20.4 s with default threadsettings in LW but when setting it to double of what is avaliable on the CPU he gained speed.

wellsichris
09-16-2010, 07:36 PM
Mr. Rid, on the scene you posted, turn off noise reduction, "this is causing the weird artifacts that look like bad AA for the perspective camera" set your AA samples to 2, and adaptive samples to .04


the bug with the noise reduction causing artifacts is something to consider, but I don't use noise reduction so this is the first I have seen this.

hope that helps,

Cheers

Mr Rid
09-16-2010, 08:03 PM
I never said that...

I shoulda re-read what you were saying. I thought you mentioned limited region.


Mr. Rid, on the scene you posted, turn off noise reduction, "this is causing the weird artifacts that look like bad AA for the perspective camera" set your AA samples to 2, and adaptive samples to .04

the bug with the noise reduction causing artifacts is something to consider, but I don't use noise reduction so this is the first I have seen this.

That does help jaggies, but I see that the Classic still renders noticeably less noise (notice the walls) in the same time. The LW benchmark scene I started with had loaded with the NR on and I didnt think anything about it.


Where is this benchmark scene? Btw, I think the benchmark scenes last came with LW8.

toby
09-16-2010, 10:07 PM
Actually the Classic camera max is 70 passes, but I dont know why use classic blur with Perspective if it renders with lesser quality AA in the same time as Classic, as in the posted examples.
Yes I forgot about PLD, because it was so bad when it came out that I've never considered using it.


Limited region optimization that Cageman mentioned is a new one to me. But the two example scenes I posted are both using limited region and I see no advantage to the Perspective camera.
If you were to create a limited region 1/8th the height of the image, you can still have up to 16 cores render it with the persp cam, the classic cam can never put more than 2 cores on it. The advanced cameras will continue to subdivide the region down to a single scanline per proc. It's a vast improvement; you'll never have an idle proc, something that happens with the classic cam all the time.



I cant think of a practical situation rendering ray traced objects and no AA so I dont know what this demonstrates.
?
It demonstrates raytracing, which, by itself, the persp cam does faster. Don't worry, I'm not done yet or claiming victory, much less am I in favor of either one being the victor in the first place. Classic is looking faster at aa, but I'm nowhere near done testing that, there may some better settings someone out there can share too.



My point here is trying to figure out why so many users maintain that the Perspective Camera is optimized to render faster than the Classic in some regard to reflections, recursions, multi-threading, and now limited region. So far, I have only seen scenes that were not faster as a coworker or forum-goer at first believed. What the user perceived as 'faster' was actually with lesser quality AA settings that the user was not comparing closely enough. Or there is some particular situation that no one has yet to produce as an example.
You've seen 2 examples of the p cam going faster and one of those was clearly better quality (capt. obvious' motion blur), I don't understand why you can brush that off so easily.

toby
09-16-2010, 10:21 PM
Where is this benchmark scene?
Right here :angel:

toby
09-17-2010, 01:48 AM
here's what I found testing the reflection-blur de-noise capabilities:

Classic cam PLD35, or Enh. Extrm
no adaptive aa
16 threads
(these are the highest settings possible afaik)

Time : 1 min 39 sec.
Notes : there's not only some grain left but there's streaks/bands in the render. Never thought I'd see this with Enh. rendering. Since it's at max. quality I'll use this render as baseline.


Persp cam
22 AA / 0.03 AS
OS 0.0
8 threads

Time : 1 min. 20 sec.

Notes : Persp cam settings were designed to match the Classic max. quality as close as possible, but they did come out a little better.
AA of 22 was sort of a 'happy valley' - higher or lower would result in noticeably higher render times. I assume this is because lower aa requires too many AS passes, and more will take too long needlessly. At AS0.02, the happy number was 25.

Oversample setting; with OS of 0.7, which I had heard was the default setting under the hood for Enh. aa, times dropped by several seconds, but made it much softer than any other of the tests. I guess the Enh. renderer does not use an OS of 0.7.

All other settings identical for all cams.

You'll notice that this is a limited region render; but it stretches
from top to bottom so as not to be an advantage to the persp cam.

If anyone has better settings let me know and I'll try them.

Capt. Obvious did motion blur, and DOF is no contest, is there anything else worth testing?

Cageman
09-17-2010, 03:23 AM
AlI I read in the LW docs about this is "More complex scenes tend to render faster with the Perspective Camera." But I cant find what constitutes a scene 'complex' enough to render a practical difference.

I would love to try out some of your settings regarding classic camera for rendering things with flickerfree GI.

It would be really nice if you could share what you usually set your classic camera to and what your settings in GI are.

:)

pixelranger
09-17-2010, 03:52 AM
This is the stupidest thread ever...

I've never worked on a production where I didn't use LW, but mostly I use Maya. And if the guy next to me came from a Truespace/poser/blender background and had a license of it I would expect he would pop into it from time to time if there was anything he knew he could do quicker there than in Maya.

Theres allways a lot of softwares used on any movie (Lightwave probably used to a small degree waaay more often that it has credit for, allthough I wouldn't expect Newtek to boast about it, or even the user, if he/she just jumped into LW to, say, quickly mergeTrigonsX, magic bevel and unwrap a model before he/she jumped back into Maya to do the rest.

Blender has a lot of nice features, as well as LW, but from a pipeline, colour management, rendering and complexity (texture sizes, poly count, hair, water) point of view there's probably NO WAY one could do it all in LW (Avatar), and if it was possible it would literally have taken at least 50 years.

I love LightWave, but as with all software, it is another pencil in the pencil case.
They all bring different strengths to the table and if I was a budding 3d artist starting out in the industry I would keep my mind open and absorb all knowledge/skills.

The people that worked on Avatar are some of the best, most talented people in the world of cg/vfx and if they used Blender you should try and find out and learn about WHY they did that instead of going "They used BLENDER?????".

Sorry but that is sooo typical n00b fanboy talk.

prometheus
09-17-2010, 04:34 AM
But how do the AA settings compare? Posting a scene is the best way to demonstrate a comparison


That was the pure volumetric simple sooty fire scene ..classic without
AA and perspective with lowest AA..at automatic (8) threads, the perspective were more than twice as fast but when changed to 16 threads, it was no change in perspective performance but classic was improve immensly.

I wonder what differences between the two camera types there is in volumetric AA if you were to render purely volumetric passes.

Michael

toby
09-17-2010, 09:32 PM
Am rendering as 16 threads on an i7 960, 3.2GHz, 12gb ram.
Here is another example scene with a buncha reflections and rays bouncing all over that are supposedly optimized by the Perspective camera.88106

Again, I get the same render time with either Perspective or Classic but with better AA quality in the Classic. There is more noise and jaggies in the Perspective Camera. Anything I do to get the Perspective Camera AA to look as good as the Classic will result in longer render times.
88110

So what am I missing? I dont see the Perspective Camera render faster than Classic Camera with equal AA result, regardless of lotsa rays or multithreading. I also find that the Classic Cam's Enhanced settings work better than any other AA settings I can come up with in LW. However I have seen that Classic AA has more difficulty with certain reflection noise.
Oops, completely missed this one.

The persp cam doesn't just look worse here, it looks horrible. Some bad artifacting behind the silver shapes. Maybe it's a result of using the 1-pass+low tolerance AS method?

toby
09-19-2010, 05:17 PM
Here is another example scene with a buncha reflections and rays bouncing all over that are supposedly optimized by the
Perspective camera.88106

Again, I get the same render time with either Perspective or Classic but with better AA quality in the Classic. There is more noise and jaggies
in the Perspective Camera. Anything I do to get the Perspective Camera AA to look as good as the Classic will result in longer render times.
88110

So what am I missing? I dont see the Perspective Camera render faster than Classic Camera with equal AA result, regardless of lotsa rays or
multithreading. I also find that the Classic Cam's Enhanced settings work better than any other AA settings I can come up with in LW. However
I have seen that Classic AA has more difficulty with certain reflection noise.

Classic camera wins this one. I can't match the quality / time with the persp cam. The artifacting was caused by that crappy shading noise reduction feature, so I switched it off for all (it never helps *anything* as far as I'm concerned, incl. the classic cam)

But it ends there, turn up the classic aa from Low to Med. and it becomes too close to call. So this test compared :

Classic cam - enh. medium aa - AS .025 - 16 threads : 5:12 min - made this the baseline for quality

PLD - 21 pass AS .025 - 16 threads : 5:17 min.

Persp cam - AA 3.0 / AS .025 - box reconstruction : 5:09 min

...with none of them being the clear winner in quality, so the few seconds difference here are pretty meaningless.
The classic/enh. cam would almost always have better quailty in darker areas, but a noisier reflection of the light against the back wall.

With max quality setting it's pretty much the same as my previous test, with the persp cam solidly in the lead. I've re-attached the benchmark with medium and high settings scenes in it. If anyone has better setting please post them!

notes :
Persp cam does not seem to have a raytrace speed advantage with recursion only at 4.
Using Oversample in this case does seem make the persp cam render match the classic render much more closely, in intensity and smoothness.

Mr Rid
09-19-2010, 10:44 PM
What I am discovering is that reflection blurring is what makes the real difference. If I render with no blurring, and just add rays and threads then I see no difference between Classic and Perspective camera speeds.

Raytrace Benchmark scene-
2000 x 1500,
64 recursions,
no reflection blur,
no AA,
16 threads- Perspective renders 1/2 second faster.
1 thread- Classic renders 9 seconds faster.
So there does not appear to be any ray or thread advantage for the Perspective camera.

After adding 20% Environment Reflection Blurring to all surfaces,
Classic - 211 secs
Perspective- 196 secs
Perspective appears to handle reflection blur faster.

Now at 1k -
Classic - 50.7 secs
Perspective - 49 secs
No real difference at 1k.

Back in jrandom's CamRoom scene, at 2k, 16 recursions, 50% noise on columns, and GI, the Perspective is only 2 secs faster. The close render time is apparently because the blurred columns represent less than half of the rendered frame.

If I copy the Column reflection blur surface to the Walls, then the Perspective camera is faster as expected, after seeing the Benchmark results.
Classic -134 secs
Perspective - 128 secs

Turn off reflection blur and increase from 16 to 64 recursions, and as expected the two cameras render the same time. Again, reflection blurring is what makes the difference.

So Perspective seems to handle reflection blur significantly faster in combination with many rays, and when much of the frame contains reflection blurred surfaces. But just having more rays or more threads does not appear to make a difference. Its also much less of a practical issue if only a portion of the frame is reflective, or when rendering at 1k or with low reflection blurring values. I also notice that with about equal render times, I find Classic Enhanced AA handling noise better but jaggies are handled slightly better with the Perspective.


You've seen 2 examples of the p cam going faster and one of those was clearly better quality (capt. obvious' motion blur), I don't understand why you can brush that off so easily.
What I am trying to figure out is the unanimous claim I keep hearing that the Perspective camera renders frames faster than the Classic because of optimization for multi-threading and 'lotsa reflections.' Obvious' example was in response to a motion blur comparison, but his Perspective example renders faster because of 'a few passes', as opposed to 66 of them, which does not demonstrate anything about multi-thread and ray optimization that I can tell. It was intended as a comparison to the baby scenes I was originally referring to, but that have different factors involved. But I dont know what the 2nd scene example is you refer to.

If you were to create a limited region 1/8th the height of the image, you can still have up to 16 cores render it with the persp cam, the classic cam can never put more than 2 cores on it. The advanced cameras will continue to subdivide the region down to a single scanline per proc. It's a vast improvement; you'll never have an idle proc, something that happens with the classic cam all the time.
I see. But thats an odd situation. I dont think everyone who claims 'the Perspective Camera renders faster!' is rendering all their scenes at 1/8th height limited region.

The persp cam doesn't just look worse here, it looks horrible. Some bad artifacting behind the silver shapes. Maybe it's a result of using the 1-pass+low tolerance AS method?
wellsichris pointed out it was due to a problem the Perspective cam seems to be having with the Noise Reduction filter.

here's what I found testing the reflection-blur de-noise capabilities:...
Cant tell what all is going on in your scene. Would you post it?

I would love to try out some of your settings regarding classic camera for rendering things with flickerfree GI. It would be really nice if you could share what you usually set your classic camera to and what your settings in GI are. :)
I dont do anything special. The more passes, the more GI flicker is leveled out. Its not 100% flickerless under very close scrutiny, it depends on the scene of course, but I normally need only 3 bounces, and finals are usually 17+ pass, although 9 passes have been enough to render the flicker unnoticeable in some cases.

My Shining elevator scene used default GI, 6 bounces, 4 recursions, and I barely got away with only 7 passes. Theres a wee bit of noticeable flicker but in that situation it only added to the filmic quality. Most of that scene also contained over 4 million polys, and about 60k PFX with voxels, most of the frame was reflective with low blurring. Since I knew it would take so long to render, I carefully compared Perspective and Classic settings, but Perspective was only taking longer to render with equal settings.

It looks to me that .6 is equal to Enhanced OS, rather than .7 as mentioned.

Mr Rid
09-19-2010, 11:28 PM
...I wonder what differences between the two camera types there is in volumetric AA if you were to render purely volumetric passes.

Michael

Photoreal blur does not work with PFX, I assume because it is yet another drawback to no subframe calculation on particles.

toby
09-19-2010, 11:42 PM
What I am discovering is that reflection blurring is what makes the real difference. If you render with no blurring, and just add rays and threads then I see no difference between Classic and Perspective camera speeds.

Raytrace Benchmark scene-
2000 x 1500,
64 recursions,
no reflection blur,
no AA,
16 threads- Perspective renders 1/2 second faster.
1 thread- Classic renders 9 seconds faster.
So there does not appear to be any ray or thread advantage for the Perspective camera.

After adding 20% Environment Reflection Blurring to all surfaces,
Classic - 211 secs
Perspective- 196 secs
Perspective appears to handle reflection blur faster.

Now at 1k -
Classic - 50.7 secs
Perspective - 49 secs
No real difference at 1k.

Back in jrandom's CamRoom scene, at 2k, 16 recursions, 50% noise on columns, and GI, the Perspective is only 2 secs faster. This is apparently because the blurred columns represent less than half of the rendered frame.

If I copy the Column reflection blur surface to the Walls, then the Perspective camera is faster as expected, after having arrived similar Benchmark results.
Classic -134 secs
Perspective - 128 secs

Turn off reflection blur and make 64 recursions, and as expected the two cameras render the same time. The reflection blurring is what makes the difference.

So Perspective seems to handle reflection blur significantly faster in combination with many rays, and when much of the frame contains reflection blurred surfaces. But just having more rays or more threads does not appear to make a difference. Its also much less of a practical issue if only a portion of the frame is reflective, or when rendering at 1k or with low reflection blurring values. I also notice that with about equal render times, I find Classic Enhanced AA handling noise better but jaggies are handled slightly better with the Perspective.

Ok, I'm temped to think that noise is noise as far as aa is concerned, but I guess we should test shadow and gi noise too.


What I am trying to figure out is the unanimous claim I keep hearing that the Perspective camera renders frames faster than the Classic because of optimization for multi-threading and 'lotsa reflections.' Obvious' was responding to a motion blur comparison, but his Perspective example renders faster because of 'a few passes', as opposed to 66 of them, which does not demonstrate anything about multi-thread and ray optimization that I can tell. It was intended as a comparison to the baby scenes I was originally referring to, but that have different factors involved. But I dont know what the 2nd scene example is you refer.
Pretty sure the only multithread advantage is the fact that the persp cam subdivides the regions dynamically.


I see. But thats an odd situation. Everyone who claims 'the Perspective Camera renders faster!' is not rendering all their scenes at 1/8th height limited region.

That was only an example of the things that make the persp cam more useful in the workflow. But it's still an advantage rendering full-frame. As I mentioned, the classic cam leaves you with idle procs all the time, where the persp cam will always have all of them at 100%. If you're rendering in passes for example, you could have many things that only occupy a small part of the frame. Imagine you have the character rendering separate, a bit far from camera, so the whole scene gets loaded up on an 8-proc render node, and sits there with 7 out of 8 procs only rendering blank space.


Cant tell what all is going on in your scene. Would you post it? I assume it will corroborate that the main factor is reflection blurring.
It was just the old lw benchmark 'Raytrace' scene with refl blur on everything.

But even if there were no speed advantage to the persp cam, which I think there is, or if it was only with refl blur, it would still win imho, because of all the other things it can do that classic can't. Why switch back to classic just to get the same speed? Even if it's 5% faster, why use 2 different cam types and risk that some of your shots will fall short because something moves too fast?

jrandom
09-20-2010, 12:25 AM
I found the worst-case scenario for the Classic camera: If you have a lot of geometry and it only takes up the bottom portion of your frame, then (due to the task breakdown structure differences between cameras) the Perspective camera absolutely trounces the Classic camera. If you're doing an animation and you have a shot structured this way, using the Perspective camera will save you a tremendous amount of time.

For the attached scene with a standard AA of 1:


Classic : 1m 20s
Perspective: 42.6s

Captain Obvious
09-20-2010, 03:35 AM
Obvious' example was in response to a motion blur comparison, but his Perspective example renders faster because of 'a few passes', as opposed to 66 of them, which does not demonstrate anything about multi-thread and ray optimization that I can tell.
Well that was exactly my point. With photoreal motion blur, you don't need that many passes because the anti-aliasing samples are mostly independent of one and other. If you still have to use passes because of things like shadow maps or certain other effects, photoreal is still better, because at least you get rid of steppiness on the actual moving geometry.




16 threads- Perspective renders 1/2 second faster.
1 thread- Classic renders 9 seconds faster.
So there does not appear to be any ray or thread advantage for the Perspective camera.
Uh, what? This clearly demonstrates that the perspective camera threads better. If the classic is faster with 1 thread and the perspective is faster with 16 threads, then that means it's better at threading.




where the persp cam will always have all of them at 100%
Weeeell... not quite. The perspective camera will never split a single line into more than one thread. So if you've got one line of pixels taking a long time to render, you can actually get stuck with waiting for just one thread.



Oh, and I also did a test with the scene I'm working on at the moment. It's still in its early stages and is rendering really quickly. With pre-cached GI and one area light, enhanced medium takes 45 seconds and perspective cam with 9 AA takes 32.

Lightwolf
09-20-2010, 03:44 AM
Pretty sure the only multithread advantage is the fact that the persp cam subdivides the regions dynamically.
Yup. And this is noticeable with large polygons, as classic multithreads the primarily visible polygons, perspective (and the others) on scanlines.

The other advantage is with polygons that overlap when rendering. In classic often there's cases where large polygons with a centre close to the camera (simple example: a ground plane) render first and then the other polygons on top of it, resulting in wasted surface evaluations.

This however is due to classic using old school z-buffer style rendering (basically a painters algorithm) for the primarily visible polygons - which is similar to how a graphics board renders.
The advantage is that it's faster with fewer polygons and many blank areas in the final image, slower with a higher number of polygons.

I.e. once you hit a few million polygons in the scene you'd be hard pressed to find a scene that renders faster in classic.

Cheers,
Mike

toby
09-21-2010, 12:38 AM
Weeeell... not quite. The perspective camera will never split a single line into more than one thread. So if you've got one line of pixels taking a long time to render, you can actually get stuck with waiting for just one thread.
Heh, well I didn't expect anyone to worry about 4 to 8 1px lines, the only way to get 100% *absolutely* all the time you'd have to have a bucket for every pixel, which ends up being slower.



The other advantage is with polygons that overlap when rendering. In classic often there's cases where large polygons with a centre close to the camera (simple example: a ground plane) render first and then the other polygons on top of it, resulting in wasted surface evaluations.


I forgot about that one. Same reason that it's recommended that you suddivide your groundplane a fair amount, otherwise everything above ground gets it's shadow raytraced, much of it needlessly; huge waste of time.

Captain Obvious
09-21-2010, 02:11 AM
Heh, well I didn't expect anyone to worry about 4 to 8 1px lines, the only way to get 100% *absolutely* all the time you'd have to have a bucket for every pixel, which ends up being slower.
Kray subdivides individual scanlines as necessary.

toby
09-22-2010, 03:35 AM
What I am discovering is that reflection blurring is what makes the real difference. If I render with no blurring, and just add rays and threads then I see no difference between Classic and Perspective camera speeds.

Raytrace Benchmark scene-
2000 x 1500,
64 recursions,
no reflection blur,
no AA,
16 threads- Perspective renders 1/2 second faster.
1 thread- Classic renders 9 seconds faster.
So there does not appear to be any ray or thread advantage for the Perspective camera.

After adding 20% Environment Reflection Blurring to all surfaces,
Classic - 211 secs
Perspective- 196 secs
Perspective appears to handle reflection blur faster.

Now at 1k -
Classic - 50.7 secs
Perspective - 49 secs
No real difference at 1k.

Back in jrandom's CamRoom scene, at 2k, 16 recursions, 50% noise on columns, and GI, the Perspective is only 2 secs faster. The close render time is apparently because the blurred columns represent less than half of the rendered frame.

If I copy the Column reflection blur surface to the Walls, then the Perspective camera is faster as expected, after seeing the Benchmark results.
Classic -134 secs
Perspective - 128 secs

Turn off reflection blur and increase from 16 to 64 recursions, and as expected the two cameras render the same time. Again, reflection blurring is what makes the difference.

So Perspective seems to handle reflection blur significantly faster in combination with many rays, and when much of the frame contains reflection blurred surfaces. But just having more rays or more threads does not appear to make a difference. Its also much less of a practical issue if only a portion of the frame is reflective, or when rendering at 1k or with low reflection blurring values.

Not done yet but PLD is beating persp by about 2x, testing brute-force gi and area light shadow noise, no reflection.


I also notice that with about equal render times, I find Classic Enhanced AA handling noise better but jaggies are handled slightly better with the Perspective.

Enh. aa might have the advantage of the Oversample, so I think it's fair to test the persp cam with that on.

toby
09-25-2010, 07:20 PM
classic enh., 70sec. at enh. extreme no adaptive

PLD, 75 sec. at 35 pass no adaptive

Persp AA 4 AS .01; 40 sec., and it's better quality

I hadn't been using the extra-low AS tolerance of 0.01 in the persp camera because it takes twice as long, but for doing tests for maximum de-noise ability, it seems to be un-beatable. At 4 aa it's smoother and faster than 25aa/no adaptive. I should go back and try it on the reflections.

Zooming into the shadow of the spheres far enough that you can see individual pixels shows the difference clearly. I ended up doing only raytrace shadows because of the absurd length of time it takes to render non-interpolated gi free of noise.

toby
09-26-2010, 02:20 PM
Classic camera wins this one. I can't match the quality / time with the persp cam. The artifacting was caused by that crappy shading noise reduction feature, so I switched it off for all (it never helps *anything* as far as I'm concerned, incl. the classic cam)
So I went back to mr. Rid's cornell box again, but the persp cam is still quite a bit slower using the extra-low threshold, and not quite as smooth. ( This is the low-quality test )

classic - enh. low, AS.0025; 2min. 56sec.
persp - AA 1 AS .017; 4min. 01sec.