About GI, RTO, resolutions and other things for animations

[Bjur]

Hi all

Wasnt really active the last months, TG wise.
Still a TG newbie, i was to start a short project last winter which should include some TG animations.
But life gone bitchy and also i got stuck in other student projects which had to be finished.

Me feels like i lost my very little knowlege and skills of TG again i learned/teached myself last year..

http://nooooooooooooooo.com/

*pushes reset button

Last time i avoided GI and its problems, right from the start and in all.
But being still at a very early stage of my project, now i decided to give GI a go - test wise.

Folks, a few questions..

GI-Cache and Frame interpolation:

Some ppl incl. Matt mentioned possible 10 frames for interpolating GI cached animations.
I guess this is just a possible value which grants a good outcome of quality and just for explaining the frame blending thing in all in the description.
But does exist a formula or "wisdom" about a possible "hero value" of GI cached frames with a good avarege and half way safe result in the end?
With pre-rendered GI-Caches you may safe some time on the one hand (in certain circumstances), but interpolation is stealing time as well i got, reading the description.
I just dont have any clue now, how much different values of interpolated frames could harm render times of GI-cached  animations.

For any quality and/or resolution, grants the preset of 5 frames the best average outcome to save some time in GI calculations inclusive a gain of quality?
Or adds interpolation for GI in every case even more rendertime, just for the hail of quality by kickin even more the ass of "Mr. Render Time" generally?
Or will i gain render time nevertheless, no matter what values i set against a frame by frame rendering?

Another question/missunderstanding:

Are there benefits of GI-Cached data for tweaking a single picture/frame (no animation)?
Thought that every tweak in terrain, population, quality or resolution etc. has to result in a new GI calculation for showing every generated content accurate.
I may have misread/misinterpreted some stuff..

Supersample Prepass and Raytracing Objects uncheked:

As i know, these settings have a direct impact especially to objects like plants, imported objects generally and so on (also fake stones maybe..!?).
But Imagine a procedural, infinite terrain/surface, without plants, imported objects or fake stones (till yet).
Does SSP and RTO still have any influence to the quality of the rest like base terrains/fractals and its distorsions, twists, warps and remaining surface shaders in TG?
In the past it seems fine for me, just to work with RTSs, leaving the rest unchecked..

I just want to safe every second of render time, even before starting a test animation, as i dont own a render farm.

For possible revelations, big thx in advance.

Alex

[Oshyan]

For GI cache blending, use 5 frames *and* use "sparse" cache, i.e. render every x frame (I suggest every 3-5 frames, unless your camera motion is very fast; if it's very slow, you might get away with every 10 frames). So this means, ender a GI cache on frame 1, 6, 11, 16, etc. (use sequence step to achieve this), then render using interpolate with 5 frame blenidng.

For static images you generally can't use a cache *while you're working on it* because, as you said, the scene changes and this should affect the GI solution too. There are certain changes that would affect it so minimally that it may not be necessary to regenerate the GI cache, but it's not that often, so not really worth worrying about. GI cache is useful for rendering large still images in multiple crops however.

RTO has no effect on the terrain. But there's no reason to leave it unchecked - if you have any objects at all in your scene, it is beneficial to raytrace them in almost all cases (except when you want to apply displacement). Supersample Prepass is for GI and is generally beneficial, equally to terrains as to objects.

- Oshyan

[Bjur]

Thank you for your fast response Oshyan

Will start a test animation soon with the preset of 5 frames for interpolating GI cached frames.

But will i loose render time by interpolation generally or will i gain render time by using interpolation?

Atm. i try to compare the outcome of renders of low quality settings in higher resolutions against higher quality settings with low resolutions - for safing render time in the long term.
To know if GI caching/interpolation will slow down or speed up GI animations generally, would definitely influence my settings before i start animations..

Maybe there is a thread out there with wanted answeres i didnt found, i´m sry then. TG seems to be still a stronghold just for stills enthusiasts.. ^ 

RTO has no effect on the terrain. But there's no reason to leave it unchecked - if you have any objects at all in your scene, it is beneficial to raytrace them in almost all cases (except when you want to apply displacement). Supersample Prepass is for GI and is generally beneficial, equally to terrains as to objects.

- Oshyan

SSP now will be checked in the future!
RTO.. Hmmnn, i dont have any objects in my scenes till yet and only my base terrain have displacements - with surface shaders on top.  My experience was, that RTO adds render time, even without objects..
I have to re-test/check RTO i guess.

Thx again.. o/

Alex

[Oshyan]

GI caching can potentially slow things down a bit. Using sparse caching has the added benefit of saving some time generating GI solutions though, so it can partly offset it. I would say at best it could be about the same with or without cache, when you consider the aspects that both speed up and slow down rendering. At worst it slows down rendering a bit, though it shouldn't be too bad. The more frames you blend, the bigger the performance impact, so I would stick with 5 (which was a lart part of why I suggested that). Also be mindful of the relative detail and sample quality because this can greatly affect the size of the cache files, and the larger/more detailed the cache files, the higher the render time impact.

For animations, the key is keeping things "stable" (i.e. avoiding flicker, of multiple kinds) between frames. This can be done at either low or high detail, but some of the same settings are needed in either case (e.g. Detail Blending) and will affect render time. You should also potentially consider the effects of video compression and post effects, depending on your intended viewing method. If you are mainly posting to YouTube, for example, you'll lose a lot of detail (unfortunately), even in the best cases. For that situation, high resolution with lower detail *may* be better. Personally I'm not sure there's a clear winner between the two, but again the trick is making sure things don't flicker, regardless of output resolution. The Animation Check button is designed to set things up so that they shouldn't flicker in most cases. Lower detail will tend to be more prone to flicker in geometry, though this *may* be partly offset by high resolution rendering. Honestly it also really depends on what you mean by "low detail". If you're talking about 0.5 detail vs. 0.8, for example, I would say 0.5 or 0.6 is probably best. If you mean more like 0.25 detail vs. o.6 or something, I would tend more toward 0.5, in other words avoid really low detail, particularly anything below 0.4.

RTO should not add render time without objects, as far as I'm aware. Definitely re-rest and let us know if that's not the case.

- Oshyan

[Bjur]

My targeted resolution/output is 1080P (thus downscaling for higher resolutions).

Test-War is ongoing between Detail 3-3.5 in double Full-HD (3840) vs. 1.5 Full-HD (2880) Detail 4 or whatever vs. Full-HD, Detail 6-6.5 (1920) till yet. :S

RTO: I will give a feedback!

[Oshyan]

Er, I hope you are talking about detail 0.3-0.35, 0.6-0.65, etc. and not actually 3-3.5!

- Oshyan

[Bjur]

Oh sh..!

Yes, i´m talking about 0.3-0.35 and stuff..! xD

[Bjur]

Hi again.

I´m sorry for my late response..

My student sidejob and university projects are still punching me forth and back, but i managed to find some time in between for further investigations and experiments of mine.

I made some comparisons between several basic settings and different resolutions vs render times vs. quality outcome the last days with TG 2.5 (GI stand alone images with animation settings).
My comparisons are based on a simple single fractal surface and a snow shader (no heightfield, no base color shader), excluding clouds, water and other objects/populations.
Both, surface and snow, are WIPs and aren't rly that great at this moment.

Here an snapshot from a 2 second test animation i made. It is a shrinked low quality image just to show one of my simple scenes i worked with for my comparisons:

[attach=1]

My experiences of double HD (3840pix) with low quality settings of detail 0.35 and AA 4 vs. Full HD (1920pix) with detail 0.7 and AA 4+ for possible animations @ GI standard settings:

To sum it up: In aspects of render timer and quality, DHD won!

A detail of 0.75 or 0.8 in Full HD wouldn't make any sense as you wont see a real gain of quality in simple scenes vs. 0.7 in Full HD
You just would have to face even more render time per frame against a DHD render @ detail 0.35.

Full HD detail of 0.7 with AA-Bloom and RTO on + the Catmull pixel filter (4.40 minutes) was needed just to come close to the quality of a shrinked DHD render with just a
detail of 0.35 and everything unchecked (beside RTS) with the standard Narrow Cubic pixel filter (4.38 minutes).
In the end the FHD picture wasn't looking better or worse, but different and needed more time to render as the DHD counterpart.

Microvertex jittering was checked in both cases.

Everything up to Full HD, the Catmull-Rom filter seems the way to go for animations to prevent blurred or muddy visuals (freakin motion blur turned off).
Using the Catmull-Rom filter for higher resolutions above Full HD didn't help as the filter adds additional render time and wont bring better images when all is downscaled to Full HD again.
So is the standard Narrow Cubic filter my 1st choice for resolutions above Full HD.

RTO should not add render time without objects, as far as I'm aware. Definitely re-rest and let us know if that's not the case.

- Oshyan

RTO (Raytracing Objects) on/off:

It just adds render time even without any objects around if checked. But not that much.
5-6 seconds in FHD, up to 15+ seconds in DHD. I guess the additional time is due to the search for objects in a scene and a part of the process itself, i don´t know.
I will let RTO unchecked in scenes without objects because there seems to be no visual quality gain at all for saving render time.

Strange thing is: Without any objects in my scene but RTO checked, RTO influenced a few little parts of the scene (fractal surface parts in shadows becomes darker a bit).
But why?

Anti-aliasing bloom:

Adds a horrible amount of render time per frame, up to 30 seconds just in FHD, with a detail of 0.7 and @ AA 4, and let fine details look even more worse than before, especially in middle-
and backgrounds when finer snowy details meets finer fractal landscape in the distance. I see there is a benefit of using it AA wise, but bloom looks generally unnatural.

Is AA-Bloom rly recommended and helpful for animations if you go with higher resolutions?

I hope its not. I am thinking about "moar pixels = less need for another AA filters or higher AA values"..
There was a visual and a time loss when using AA-Bloom. It may be helpful in special situations i don't know or in animations generally in lower resolutions. Tell me!

GI:

All seems fine with the standard settings of 2, 2, 8. There was just a little bit of a quality gain when rising the sample detail to 3.
It resulted more in facing too much additional render time in very high resolutions.

But i switched GI atmosphere illumination to 0, which safes me another holy 1+ minute.
There was no visual gain with GI atmo @ 1. This may change if clouds come into play i guess..

*Sidenote: All called times are based on a 2700K @ 4,6 Ghz. My silent renderslave isn't that fast as it is just a single Xeon 1240v2 @ 3,6 Ghz.

The only thing i am rly afraid of using DHD clip sequences is the fact that 1 DHD image is 17,5 MB big (a FHD image needs just 4-5 MB disk space)!
Imagine 10.000 or whatever frames, i will have to buy additional hard disks and i dint know if my hardware could proper handle a kind of 4K workflow.
Would also have to buy more RAM i bet. In the end there is chance that DHD clip sequences could stomp my complete pipeline workflow into the ground.
Did anyone have experiences with "4K" clip files when using Nuke for postproduction. For Maya relevant work i could use downshrinked alias clipfiles.
This may work with nuke too, workin with alias clipfiles and for final renders just switch the source material footage, hmm..

Additional questions:

Why does my into the background projected HQ .tiff image (the lil planet in my animation) is looking so ruff, even with high render settings?
Because of this, i had to soften the outlines hard in Photoshop. The quality outcome, as you see in my non GI animation, was still mediocre.
Commin atmosphere samples into play, if something is projected into the background for a better projection quality, huh?

For better average quality, especially with lower AA values like 4 (4-16), is to check for non-adaptive max. samples recommended?

Additional suggestions of a beginner:

Visible sun:

Very cool would be a slider for independent sun settings for its visible size and halo/glow amount, which wont have any impact/influence to the given light
conditions of atmospheres and surfaces (suns strengh would/should still dictate the atmosphere/surface lightning).
With this we could say "kthxbye" to additional fake suns and other fake solutions which are just placed because of haze and blue sky outrages for providing
a better visual outcome in the case the sun becomes/is a part of visible content.
Just imagine a "good" scene/lighting in animations, which will not get destroyed by "Here comes the sun.." *sings.
Stand alone image fans would be pleased too i guess..

Animation helper:

Also nice would be a kind of an embedded "frames vs. moving range" calculator/warning, which grants a minimum of a flowing animation generally close to surfaces if checked.
You can set 10000 frames for 10 meters, no problem. But if faster or different movings comes into play, it would be great to know you may need more frames for a flowing
animation for your chosen distance.

Motion blur doesn't help in all as it just fakes the impression of movement and cant fix missing frames.
For a wanted 48/50 frames instead of 24/25 frames for example (there have to be a base value near surfaces of course) you would just need to raise the recommended frame amount x2 then.

Everyone who needs speed could speed up movements via post if needed, still having a "granted" flowing movement without any shutter/jerking.
Everyone else with special ambitions could tweak frame rates and outcomes on its own..
It would also help to calculate render times better right from the start (for long or lots of animations) and could prohibit wasted test animations for every shot aka "Was it flowing now?".

Had to render my old, end 2012 1st test animation finally with 30 FPS, intended 25 FPS were too choppy, even my imagined spaceship/camera wasn't flying that fast.
Sometimes it felt that TG eats some frames resulting in feel of 12 FPS. This belongs of course more to foregrounds and when close to surfaces.

Last but not least - The "Rule them all" button: Check it and all your suns and planets are circular and wont look like eggs..! ^^

At last: My very 1st TG work ever posted/shown till yet! *Fanfare*  *hides in shadows

Both versions, on YouTube and on Vimeo stutter/are choppy a bit. Source materials, even compressed ones are way more flowing (and have higher quality of course).

[vimeo]http://vimeo.com/65726318[/vimeo]                                                   

I decided to make my 1st test animation with an old scene from end of 2012!? i found in old project folders in lower resolution.
Thought it would be better to see if HD works with actual settings instead of not ready tweaked actual scenes/landscaper in very high resolutions.
It worked. No flicker, shadow or fractal surface popping! And when it looks good in HD, then FHD or DHD will do i bet. Next step would be the same animation but now with GI.
This next animation will be posted here too because of my started topic. For later animations aiming at final settings i would open a (guess neverending) wip thread
in the animation section, which needs more love to be honest - or more breakdowns/clips of some industry pros out there like that one of chris_x422
about his TG work for Snow White for example.

Greetings, Alex