TG3 vs TG2 raytrace atmosphere rendering speed

Started by DannyG, February 05, 2014, 12:34:49 PM

Previous topic - Next topic

DannyG

I have to agree in regards to the Defer atmo being unpredictable compared to v2's Raytrace atmo, With v2 - Atm @ 16-24 samples and any Large Cumulus set @ 64-96 Raytraced w/ AA 8 or more would normally render clean. The obvious sample changes in the cloud nodes are certainly throwing me off a bit. I have yet find a setting that I am comforatable with to use as a "go-by" for renders. This topic came up at the v3.1 presentation Oshyan gave last week, Matt touched on this a bit mentioning that the samples and some other changes might be the cause, however because of the topics being focused on at that time he was unable to give a detailed answer. For now like Martin I tend to stay away from Defering.Thats my 2 cents
New World Digital Art
NwdaGroup.com
Media: facebook|Twitter|Instagram

Tangled-Universe

Hi Danny,

FYI...Ray traced atmosphere and Defer Atmo are the same thing. You shouldn't find much difference in defer/rt atmo therefore, unless Matt changed a few things under the hood for TG3. In essence they are the same.

DannyG

Quote from: Tangled-Universe on February 05, 2014, 03:55:59 PM
Hi Danny,

FYI...Ray traced atmosphere and Defer Atmo are the same thing. You shouldn't find much difference in defer/rt atmo therefore, unless Matt changed a few things under the hood for TG3. In essence they are the same.

I guess I would need some clarification or correction. I personally do see an increase in time and inconsistency from scene to scene with defer compared to old Ray trace. Perhaps something was changed under the hood as you mentioned.
New World Digital Art
NwdaGroup.com
Media: facebook|Twitter|Instagram

DannyG

#3
File was created in v3 than opened in v2 and rendered. No settings were touched between applications. Soft Shadows are enabled and set to 5. NO GI in either render
New World Digital Art
NwdaGroup.com
Media: facebook|Twitter|Instagram

Oshyan

#4
Thanks Danny. I'm testing your file now and would like to split this discussion out into its own thread. I'll do that once I have some tests done. By the way though, when testing between major versions like this it's almost always better to create the scene in the earlier version (TG 2 in this case), then load it in the later version, because TG3 has features and settings that TG2 does not, and TG3 is backward compatible while TG 2 has no knowledge of the existence of TG3 or its settings. So what you'd ideally want to do is create the file *in TG2*, then test in TG3. That being said it doesn't necessarily explain the results you're seeing.

I also want to mention that your settings are, well, a bit insane quite frankly. AA8 with Defer Atmo/RTA is seldom advisable, but *especially* with 24 atmo samples and Quality 1 for clouds. It's no wonder that Defer/Raytrace is so much slower here. Oh and there *is* GI since the Enviro Light is enabled and GI Det/Qual are 2 and 2. Perhaps you meant GI Surface Details? But that doesn't affect the atmosphere at all anyway.


Edit: these posts have now been split off into their own thread.

- Oshyan

Matt

#5
Sampling changes were made for TG3 which increase render times whenever you render with soft shadows, but they allow much higher quality soft shadows in difficult cases that weren't possible to render before. Until now, my suggestion had been that if you run into speed problems with soft shadows in TG3, you can reduce atmosphere samples and/or cloud quality because the TG3 results would be higher quality with the new sampling methods. In TG2, enabling soft shadows would often make the atmosphere/clouds noisier so you needed to increase samples, while in TG3 this isn't the case. That means that if you have soft shadows enabled, you should use lower samples or cloud quality in TG3 than in TG2.

However, I'm testing your scene and TG3 is quite a bit slower than TG2 even with soft shadows disabled, and at this point I don't know why. I'll investigate further.

Matt
Just because milk is white doesn't mean that clouds are made of milk.

Matt

#6
Hi Danny,

I found the main cause of the difference between TG2 and TG3. It was some code that was added to support render elements (specifically the cloud 2d motion element) but it was being calculated even when you didn't need it to. It affected all atmosphere and cloud rendering, and even the shadows of clouds and atmosphere. It doesn't matter whether you have Defer atmo/cloud enabled or not, but Defer atmo/cloud greatly increased the render time in your test, which exaggerated the difference between TG2 and TG3.

I've correct this for Terragen 3.1, which will be released today. Renders that spend a good amount of time on the atmosphere or clouds but don't output the cloud 2d motion element should render a lot faster in 3.1 than they did in 3.0. Renders that do output the cloud 2d motion element will also be somewhat faster because 3.0 was wasting time calculating motion information in shadows when it didn't need to.

In some cases I found 3.1 renders slightly faster than 2.5. This may be due to various other minor speed boosts we've accrued along the way since 2.5. Terragen 3.0 -> 3.1 has some small optimizations so these might be helping too. I don't know how generally this will apply, but it's a positive sign anyway :)

Matt
Just because milk is white doesn't mean that clouds are made of milk.

ejgodwin


DannyG

Yep, huge difference in 3.1.02.  Thanks for the hands on Matt ^^
New World Digital Art
NwdaGroup.com
Media: facebook|Twitter|Instagram