190h05m and still rendering

Started by TheBadger, February 28, 2012, 09:53:25 PM

Previous topic - Next topic

dandelO

#30
I really want to see your finished scene too, Badger, it looks great already.
On the rta thing, most of you'll know I'm a real render time miser due to a pretty low end computer so I always look extensively for the fastest render method. I find that rta is always faster than not using it BUT(big but) it really depends on lowest sampling levels of AA. This means making use of the adaptive sampling at render time.

Where TU(I believe) is absolutely right that his experience has proved quicker by not using rta, I think you must be using either default or max samples at render time for AA. This will certainly get into longer render times the higher AA you use because the minimum samples per pixel do begin to increase more quickly/drastically after AA=4, indeed. I like to drop the 'minimum first samples' and keep a reasonably high cloud detail and usually never below detail='1'. That should ensure that your cloud is nice and smooth at most settings and you will be able to use, for example, AA=8@1/64th sampling, to give a minimum 1 sample per pixel(a lot of samples isn't needed because the detail is already there from the atmo/cloud quality) and render time is managed well and fast.

The other way around, that I think Martin likes to use more, is to keep AA sampling either default or maxed out and use a lower cloud quality, because not so many samples are required at higher AA levels. The more minimum samples you throw at it, the longer a rta rendered atmosphere will take.

Popular advice and recommendation says that using higher AA and lowering cloud/atmo quality to balance is the best way to use rta. I think, contrary to that, that I can usually always make an atmosphere/cloud render faster with using rta than without it.

In a sentence, keeping a decent cloud quality and lowering AA minimum samples is faster to render than higher AA samples and lower cloud quality. And usually always of higher output quality than a non rta rendered scene.
That said, it's all scene dependent, sometimes you can't get away with lowering the AA samples.
I definitely prefer to use rta when I can because it is generally much faster, for me, in a typical scene.

TheBadger

I need a small clarification on a definition.

Atmosphere: as it relates to Terragen2.
We are or are not, simply talking about the *sky*?!
We are or are not, talking about all the space reveled by light in a scene, that is not terrain?

A simple but important distinction I think, and I'm really not sure of the answer. Please remember that TG2 is still the only 3D software that I am familiar with, so I need to ask.
It has been eaten.

TheBadger

Ok, so render two is at (right now) 37h6m and nearly finished, much faster than before.

Again the render is having a hard time finishing at the bottom left corner.

Most of you who are talking on this subject are in or near Europe (I thought). I am in the midwest US. So I will wait to stop the render until tonight, if it has not finished. Since you are all probably sleeping right now anyway ;D.
I can tell you that there is a clearly visible quality difference in the two renders; Long vs. short render, way more than I thought was being suggested.

both images are at 2500pix wide, so it should be easy for everyone to see the differences. I'm sure it will help you guys to pin this matter down ;)

See you soon.
It has been eaten.

TheBadger

#33
Image 1

320+hours quit before finish
2500pix Wide
RTO-YES
RTA-YES
GI Surface Detail-YES
detail:0.9
AA6
GI 3:3:8
DO Soft-Yes 1:11:1

Image 2

Finished at 45:24:15
2500pix wide
RTO-YES
RTA-NO
GI Surface Detail-NO
DO Soft-Yes - Default
Detail 0.75
AA6
GI 2:4:6

There are, I feel, very noticeable differences in the two renders.
In both Renders there was a long hang in the render finishing, when the render got to the bottom left of the images. It work on a tiny part, for many hours.

These images are WIP(s)
It has been eaten.

dandelO

Very nice indeed. I love the first one, such a shame that GI surface details takes up so much time because that's what's making image 1 really sing, the GI in number 2 seems flat in comparison. Great looking scene, though!

Kevin F

Great image and I agree with dandelO about the quality between the two.
So now for the fly over?  ;D

Dune

A little post in Photoshop will enhance the second render a lot. I think you can make it almost look like nr 1 with a little effort. Because 320 hours for only 2500px wide image is quite something.

Oshyan

Badger, great to finally see the scene. I think it looks really nice, but yes there is definitely a big difference with the GI Surface Details. This *may* be one of those rare instances where it's actually "needed". The many small ivy leaves are the cause of it, essentially. That being said I'm quite curious to determine if we can at least get closer to the GI SD result without quite so much render time. The high render time even without GI SD leads me to believe there may be further room for optimization, though with the large object(s) size, small compute terrain patch size, and very small leaf size, perhaps I am being optimistic. It may be a moot point anyway as it takes so long to render on your system, I wouldn't expect you to do more renders just for testing, and even if you were willing to share the scene and assets, they are so large it might be impractical to transfer them (depending on your connection speed). What are your thoughts on where to go from here? You said you had some more questions?

Martin (T-U, not dandelO ;) ), I think we're both right here. We're both speaking from our admittedly subjective experiences. dandelO is right too, from the same perspective. It is clear at this point that RTA can be both good and bad, depending on the scene specifics and the other settings used. Now, can we produce absolute, scene-independent guidelines on this? That's a much tougher question, and I'll agree with what you might say next: it probably shouldn't be (or at least it would be nice if we could). If it's that difficult to provide good, consistent rendering setting advice, then ideally either the settings would be made more automatic, or easier to use at least. It's something we do work on all the time, but it's a tough problem to crack. I think you'll find most renderers have lots of obscure settings that you need to understand both in function and in *practice*, across a variety of scenes, in order to fully utilize. Just a glance around at render setting discussions on Vray, Vue, Maya, and many more shows this to be true. So perhaps the most realistic goal, then, is to have a renderer whose base settings have the biggest impact on render time vs. quality in the *majority* of scenes, and the more advanced tweaks are much more seldom necessary. It's a work in progress (including the docs)...

- Oshyan

rcallicotte

Thanks Badger.  Great information to think about and excellent renders. 
So this is Disney World.  Can we live here?

inkydigit

really great results!... some optimization tests sound encouraging also!
:)

dandelO

Quote from: TheBadger on March 08, 2012, 03:38:59 PM
Atmosphere: as it relates to Terragen2.
We are or are not, simply talking about the *sky*?!
We are or are not, talking about all the space reveled by light in a scene, that is not terrain?

Sorry, missed this reply, Badger. When I say atmosphere I mean everything from cloud in the sky to the volume of the haze, which envelops the planet(I would dearly love to see an atmosphere node volume control that would accept shader inputs, wouldn't that be cool? But that's another story) For the best definition of atmospherics and all it relates to is to visit the default render and look at the difference between two renders, one with 'atmosphere visible', one without.

FrankB

what a great render and scene, I totally love it! I've made a crop to fit this on my desktop (and adjusted color etc too a bit). Here it is attached.

Cheers,
Frank

elipsis1

Stunning!  Excellent job :)

I need to read the entire thread, but was this a heightfield with added population objects for the plants?

wow, I'm just stunned!

TheBadger

@ JO
If you are reading this, would you mind sharing what you would have your *advanced* settings at in the render tabs. Perhaps I can get some speed gain there? What would you have all those settings at if this was your project?

@Oshyan
I would be willing to send the project and all related files to you if you think you can make it work better. I do want the quality of the first render with a render time closer or better than render 2. Lets think of a way to transmit the file incase thats what ends up having to be done; my connection speed is fast, but for more than 2gb it going to take a bit no mater what.

As for right now I am resigned to doing crop renders of any changes I make to the shot and using photoshop to put them together.

@elipsis1, Inky, calico - thanks
@DandelO- Thank you for helping

@FrankB-
Hi, thanks for the complement. If you really like the concept here, I will make sure you get a high quality jpg when its finished.

I'm not ready to give up on this project just yet.
It has been eaten.

Kadri


TheBadger  i really like the first image :)

But before i posted here i did some tests. I can not believe that the GI Surface Detail-NO  makes the difference (so much).
It looks (after some tests) to me that you probably have some different objects settings.
I think the second image do not have any ivy leaves shadows .
Are you sure you did not accidentally unchecked the objects "Cast shadows" settings or anything like that ?