RTA/Non RTA/customised sampling + Render Comparison Outputs

[dandelO]

I knew I'd saved some test renders somewhere about this, so after some hunting I found 'New Folder(2)' buried among the strewn wreckage of old projects.

With lots of debate and talk recently about the benefits/hindrances of using the raytracer for atmosphere rendering, here's how I use the customised AA sampling methods to get the most speed out of the renderer as possible.

Firstly, here's a render of a default cloud layer, without RTA.
Cloud quality='1', render detail='0.7' and AA='3/default sampling'. A quite respectable level of detail, so no noise. 9 minutes 1 second.
A long render due to this computer but that isn't important, what is important is this render time, relative to the following RTA versions in the next post...

[dandelO]

And here's all the RTA versions. I think I'd rather render the ones at around the 3-5 minute mark, the gain in quality over time isn't worth it, in my opinion.

A really simple example that just shows basic clouds with minimal noise but it is interesting to notice how the render times relate.
In this instance, I actually like the look of the non RTA one in the first post best but for sometimes under half the time of that one there isn't really that much drop in quality and sometimes none at all.
And, the differences between different settings is isn't generally available to your viewer so they don't have the reference points to judge your actual output by.

The mumblings of a render time cheapskate. ^^

[rcallicotte]

Animation diminishes the need to have the quality at ultimate, so I like this idea - sorting out render times to show the small differences.  I don't see much difference between AA2 / Render Quality 1 vs. the longest render in the bunch.

[TheBadger]

It is hard for me to see it too. DandelO, would you mind pointing some of the differences out and sharing more of your thoughts on them? I see your conclusion but the evidence of your premicess is to subtle at such a small size.  You know your work better than anyone. Also, thanks so much for sharing your tests!

[dandelO]

It was just something I done a while ago to try and work out how to get comparable quality in a quicker time range by using rta with customised AA sampling for speed.
I think you know already the conclusion, just that although all the images in the test are of roughly equal quality in visual output, they vary massively in relative render times. It shows that sometimes using really high settings isn't always necessary and can be very wasteful of render time with really no remarkable improvement for all that extra time.

So, pointing out the differences is just a case of saying look at the final render times for each image in the list, they all look very similar in terms of quality so, you have more time to play instead of waiting. If you manage to get good results in half the time, even quarter of the time, it's all good.

[dandelO]

I suppose, another thing it shows in one case is also that using a lower final AA number(the one with AA=4@max samples) is a significantly longer render than the next one at AA=6@default sampling.

There's really loads of room to juggle and squeeze render time by using good combinations of the many available options for controlling quality, throwing loads of high quality settings won't always be the best use of my time, or even make the results significantly better in lots of cases, so I look for shortcuts everywhere I can.

[airflamesred]

Good work Martin

[Tangled-Universe]

Bedtime now, so I think I will look at this tomorrow in more depth, but judging quickly I think these are nice tests. However, the quality setting of clouds should be considered kind of similar to the detail setting in the renderer. Once switching from rasterized rendering to raytraced rendering the quality level setting is less important and the visual quality and render speed is mostly determined by number of samples. I feel.
I believe there's also a relation between the depth/density of the clouds and the performance of RTA vs rasterized renderering, in favor of rasterized rendering.

I'm pretty sure you've seen and understand my tests here:
http://forums.planetside.co.uk/index.php?topic=13679.15

[jo]

Hi,

Something I noticed from my quick tests with basically default settings and RTA on/off is that while RTA-on was normally a bit slower (this really is with default settings and a 3D cloud layer or two added) it did look a lot better. When you compared the two you could see the micropolygons in RTA-off quite clearly whereas RTA-on looked much better. As it was default settings we're talking about renders on the order of 1-2 minutes so while proportionally RTA-on was much slower in real terms it wasn't a big deal. I didn't try it but I think my naive attempts to get RTA-off matching in quality would have easily brought the render times closer together. When I tried it with more complex scenes, again only turning RTA on and off, RTA-on became faster and better looking.

That's unscientific and anecdotal, but there you go anyway :-).

Regards,

Jo

[dandelO]

I do remember you posting those, Martin. I also remember some other rta comparison you made a while ago showing how to get cloud cloud quality down to something like 0.02(or something else ridiculously small) and it still looked good too. That's kind of the other way around than I've done here, by keeping cloud quality at 1 and reducing the AA samples. There are times when using too sparse AA won't work well this way but this just showed me that I could get a lot of similar outputs with times ranging between 3 minutes and 22 minutes with very little difference to look at. Just an experiment from a while back that I thought might be handy for anyone reading who might like to mess around with the customisable AA a bit. I think, as with most things TG, there's multiple different answers to any given question or issue, the program is so open and it lets you control pretty much anything, this is just another way that makes good render times, for me.

[dandelO]

The option to control the adaptive AA is a really great thing. A good way I find use it is to have cloud quality at maximum with RTA and an AA value that allows for 'min samples per pixel = 1'.

[FrankB]

To add to that, I found another situaltion where RTA on is more efficient that RTA off.
When you render a huge cloud that would take you 500+ cloud samples with RTA off to render without noise and lots of detail, you're actually better of using RTA on and a muuuuch lower number for the cloud samples. Usually AA=3 suffices, rarely you'd go to AA4.

Frank

[rcallicotte]

I still think that high quality settings will matter much less if using TG2 as an animation tool.  Is this possible in the future that TG2 will become more of an animation tool than an export tool for games / movies?  Or just for still visuals?  Animation seems to me to change the entire field of fun.

[Kadri]

Not sure if i understand you right Calico but the problem is that high quality settings will matter much more when you make an animation.
You can get away with some noise in a still image or bad rendering parts , shadows etc.
But when that noise changes from frame to frame and you get a lot of flicker the fun begins to make your life suddenly unfunny unfortunately
The problem is to find the balance between speed and quality.
And as Osyhan said in another thread many renders do have many options for rendering.
I would love to see a basic slider with settings like "Still image -low-medium-high- quality" with an  "Animation" box as needed.
For ordinary people this might be enough maybe but the Professional crowd would like the deep options more probably anyway...
I remember Matt said in an old post that he doesn't like so many options too. I am curious if he will get them easier to use  

[FrankB]

In this discussion, you should probably best distinguish between "quality" and "detail".
Depending on the animation, you could for example get a away with a lower geometry detail, but you still can't have bad quality, such as noise and flicker.