GI Cache files

[Henry Blewer]

I rendered the GI cache files, one file for every 5 frames. The GI Surface Details was checked, settings 2,4,8.

Since all the GI has now been rendered, do I have to leave the GI Surface Details checked? The animation frame renders would calculate faster with this un-checked.

[Oshyan]

A: yes, B: don't use GISD for an animation. Please. :-p

- Oshyan

[Henry Blewer]

Starting animation over. Thanks Oshyan!

This will speed up things considerably.

[Tangled-Universe]

Yes it will definitely speed things up.

However, I'm not sure about Oshyan's answer for A.

As far as I know GI surface details (GISD) is calculated during the render itself.
GISD will shoot additional ray(s) at a point to gather more light from nearby surrounding surfaces, thus improving surface detail of lighting.
Hence, it's name.
Since it's an at rendertime effect, GISD is not included in the GI cache.

So my answer would be "no". Uncheck GISD and use the caches you already have.

I would wait though for official confirmation or dismissal of this

[Tangled-Universe]

Ah found it, seems my memory still serves me...

http://www.planetside.co.uk/forums/index.php/topic,15890.msg154795.html#msg154795

[Henry Blewer]

I am using AA 8. The atmosphere has RTA on with 14 samples. On still renders, this has seemed to be fair quality wise, and not too demanding time wise; generally about 17 minutes for an image 1920 x 1080 using 8 threads.

I used GI Detail Sampling to calculate the GI cache files. 2,4,8. I thought that pre calculating the GI would cut the render times with this on. It did not.

I am now re-calculating the GI cache without GI Sampling turned on. The render times for each frame should be about 1/3 of the time required before.

[Tangled-Universe]

Did you actually read our replies Henry?
It's no problem if you say our posts weren't really clarifying to you. We'll just go over it again, no problem.
Reading from your reply you're at least mixing a few things up and I'm not sure what you're really doing at the moment.

So just to rephrase:

GI surface details is not included in the GI cache file.
GI surface details will be calculated during the render after GI pre-pass has been calculated or the GI cache has been loaded.
With animations, don't use GI surface details -> mind-bogging slow.
Generating GI caches for animations will speed up your render, unless you save out a cache for every frame.
So why did it not give you a speedup?

I'm extremely suprised to see that your frames render so quickly with AA8 and 14 atmo samples with RTA.
Makes me even wonder what render times are with 48 atmo samples without RTA.
RTA is very expensive on atmosphere, even more than on clouds.
Did you test this thoroughly enough?

Can you explain about the GI in accurate parameters names, because I don't understand what you're doing?

[Henry Blewer]

To be honest, I don't quite understand this subject.

I am rendering every five frames for the GI cache. The renders will use the interpolated method for every five GI cache files. Detail setting is 0.65 with AA 8, RTA on. Atmosphere samples 14.

The render pixel sampler's First sampling level is 1/16. so the Min samples per pixel is 4, instead of 16.

I had GI Surface details turned on. I turned them off.

How I understand it, these settings now basically give me something like your suggestion of AA 4.

It has been my experience that low sunlight angles have much more noise than higher sunlight angles. I am making a sunrise animation.

Anyway, I'll have the GI cache files calculated in about 9 hours. I'll be able to see how long each frame will take to render after work on Wednesday. It should be much faster with GI surface detail off.

[Oshyan]

I wasn't sure, but was assuming he wanted to know if he could get the *benefit* of GISD from the cache without leaving it checked. That seems to be the case, hence my answer. But yes Martin, as you noted, GISD results are not stored in the cache.

Most of Henry's response makes sense *except*:

How I understand it, these settings now basically give me something like your suggestion of AA 4.

I'm not sure what this is referring to, nor how any of the settings mentioned are equivalent to "AA 4".

If you're getting render times as low as you say Henry, your settings seem fine, so just go with the disabled GISD and have fun. As Martin said, you don't have to wait for new cache files to render, but you may already be done with that. Looking forward to seeing the results.

- Oshyan

[Tangled-Universe]

Most of Henry's response makes sense *except*:

How I understand it, these settings now basically give me something like your suggestion of AA 4.

I'm not sure what this is referring to, nor how any of the settings mentioned are equivalent to "AA 4".

Hi Oshyan,

I think he's saying that because he's possibly following my opinion that RTA is generally economic up to AA4.
With higher AA things get slower or drastically more complicated to get rendering as quickly compared to non-RTA with sufficient atmo samples.
I have stated that a couple of times before in different topics, so I probably have to plead guilty here

He's right that AA8 @ 1/16th has 4 minimum samples is equivalent to default AA4 @ 1/4th, but only if you consider minimum number of samples(!).
So perhaps in this case taking 14 atmo samples (that's non adaptive) with minimum of 4 AA samples gives the result straight away, without needing further AA.
That depends of course on the Pixel Noise Threshold, which changes to 0.0375 if you go from default AA3 to AA8.
So it seems 4 AA samples will result in a lower noise level than 0.0375 (whatever that number means, I suppose it's luminance based AA).

Perhaps you have hit the sweet spot here, but as I've said I'm a bit skeptical.
However, Oshyan is absolutely right. If it works with these nice rendertimes then just proceed and don't care

Anyway, Henry, it seems you get it to work now

Looking forward to the result too!

[Oshyan]

Oh that definitely explains what he meant Martin, I understand now. It's just that AA4 wasn't referenced anywhere else in the thread. Thanks.

- Oshyan

[Henry Blewer]

I am getting a rendered image about every 33 minutes now. This is using 6 threads. I thought it would be quicker, but I am using large populations of high poly count  trees. (didn't really need to use the high poly trees) This is still half the former render time of the first run.  It will take about 6 days to render the animation now (unless I stop to fool around in Skyrim)