Just curious: do you guys use volumes a lot? I almost never use them, except when I happen to need some clouds in the background, or a puff of smoke. 
I know it can be used for atmospheric foggy lighting, but to me thatās more of a gimmick, not something I use often, also because of the increased noise / rendering times.
Yeah, itās very rare in my workflow to have any volumes, let alone nested volumes.
Not a lot but Iām doing a personal project where Iāve decided to use them as learning something on this.
Actually, in my daily job (unfortunately, at the moment, not done with Blender) I donāt use at all (or really really rare)ā¦ so if I switch to Blender it will be the same.
In any case, in my opinion, if you do environments renderings, volumes will be really helpful to improve output qualityā¦so, for those people, some regressions on rendertime will not be so good to have.
Yeah itās a gimmick until itās notā¦ Do you use camera tracking often? I donāt. We all do different things.
Right. Null scattering was meant to make the fix less hacky.
Well, I use volumes all the time for archviz and environment/ landscape design. The old algorithm works fine, but VSP + Null scattering is definitely better. Denoising volumes is always tricky, so better convergence really makes a difference.
no and itās because itās slow to render, otherwise iāll be using them more
I donāt see an issue in removal of Volume Steps.
Volume Objects and Geometry Nodes are allowing us to control volumes resolutions and volumes merging, to obtain desired quality.
So, Volume Steps seems redundant, now.
To gain in performance, lowering amount of voxels is a lot more effective.
And the current situation is very far from what I experienced few years ago ; when any attempt to render volumes were requiring dozens of minutes.
With volume path guiding, we can have few volume bounces without a render time taking ages.
Nowadays, people are complaining for less than a dozen of seconds.
After reading all the bad things about the new Null Scattering, I thought Iād give it a quick try with a production scale VDB ( https://disneyanimation.com/resources/clouds/ ). I used the largest version of the cloud which comes in at ~2,9GB.
I loaded it into Blender, added a Nishita Sky, set the noise threshold to 0.025 and the volume bounces to 2 and left all the rest of the settings at factory defaults. BTW I didnāt aim for a close match with the cloud renderings on the Disney site but just slapped together a VDB and a sky texture 
My system is a 5 year old XMG laptop with a RTX 2070 Super Mobile with 580.65.06 drivers on Linux.
I started in Blender 5.0, current build of main. The F12 render starts in less than a second and after 3:52 minutes this is the undenoised result:
When checking the āUnbiasedā option the rendering took ~10 seconds longer, but the both renderings were visually identical.
Then I did the same in Blender 4.5.1 LTS. I made sure to set the sky model to Nishita, because otherwise I wouldāve gotten a black result (because Cycles then wouldāve used the outdated and now removed Preetham model.). Apart from that, same settings, same system.
F12 render takes ~15 seconds before the first rendered pixels appear in the viewport. Then with the default Render Step Rate of 1 I get this:
You might wonder why itās so noisy. Itās because after 3:50 minutes Cycles was at sample 30 (instead of 4096 in Blender 5.0 at the same render time). The remaining time reading was over 8 hours, which I didnāt want to invest. So I increased the step rate to 10, to 16, to 24 and finally to 32 to reach a finished rendering with 4096 in 3:50 minutes like in Blender 5.0.
Now the result looked like this:
Noisewise it looks similar BUT there are ugly stripes like these
and compared to the Null Scattering render the cloud is now missing details because of the increased step rate (4.5 compared to 5.0):
So for me not everything is worse than before but quite the opposite. Especially not having to fiddle around with step rates to find the sweet spot between unbearable render times and ugly renderings is a benefit in this example. YMMV though.
EDIT: Increasing the Transparent Bounces to something like 128 got rid of the stripes. But again another setting to remember to get good results with the āoldā volume rendering.
For atmosphere, I typically use a big volume with a very low density, light blue color and toggled only on camera rays so it doesnāt impact rendertime too much (or at all, really). This is preferable to adding the atmosphere in post, because it doesnāt play well with the depth pass when you render with depth of field. Thereās probably a solution Iām not aware of, but I am happy with this strategy so far.
Then thereās clouds, and finally absorption in bodies of water (I have a lot of ocean in some scenes). Thatās it. But these volumes are typically homogeneous, thereās no textures involvedā¦ itās an easy use case.
I use it quite often for atmospheric foggy lighting.
It can also be very useful for quickly hiding stuff, like too abrupt transitions between objects and stuff like that.
Large volumes that encompass the entire scene are quite useful for tying everything together even if it is just a slight, barily noticable fog.
The render times donāt increase that much if you donāt use complex shaders or complex imported VDB files.
It often saves a lot of artist time at the price of somewhat increased render time.
Cool, thanks for the test. I recently wanted to include a couple of VDB clouds in a project in Blender 4.5 but ultimately had to ditch dem due to render time issues. Great to see that Blender 5.0 tackles this problem.
How do you use volumes this way ? can you give an example ?
Recently I had a scene with a forest in the background. Snowy ground and trees sticking out of it. Now, the trees coming out of the snow allways looked unnatural because the the transition from snow to tree was just too abrupt.
The solution would have been either creating a snow pile around each tree stem to break up the uniformity or to create some sort of snow to dirt transition around the trees. However, it were hundreds of trees and a simple height fog at ground level hid the problem nicely and took like 10 minutes to implement.
Great to hear 
As I said, your mileage may vary but so far Iām very happy with the new Null Scattering. And I used this behemoth of a single cloud as a test case to see how Cycles deals with it. Actually I had more problems with the viewport becoming laggy (before setting it to only display points instead of the volume itself) while Cycles just rendered without a hitch.
Copying the Volume object 10 times and randomly distributing it across the scene caused a whopping 200MB increase in render memory usage, which is awesome and allows me to fill my hole environment with dozens of killer clouds like this without even coming close to the GPU memory my old laptop GPU has (8GB).
And I can even stack these clouds inside each other without risking visual artefacts.
Have you tested how it holds up regarding flicker when rendering an animation with denoising?
Did you try reducing the volumeās max steps? Itās one of the biggest optimizations you can do in the old system. The default of 1024 is crazy high, you can cut it in half multiple times and not see any visual difference.
āiād rather haveā is kinda the point. different projects and different users have different needs. youāve cherry picked an example with multiple intersecting volumes that does cause an issue that the new system resolves, but there are absolutely cases where the visual differences might not be meaningful to the artist and they can literally cut their render times in half or more.
just in case people are taking me the wrong way, i am not against the new method. iām positive it will come in super handy! i am just expressing that exclusively offering the new method for sampling volumes does have a meaningful performance cost, not a minor one.
my opinion is just that this is one of those cases where ābothā methods should be kept, at least until the performance penalty can be shaved to consistently no more than 10-20%
No, I havenāt yet. I know this setting and Iām sure I can spend several minutes / hours finding the optimal setting that renders faster than the new system.
But thatās exactly what I donāt intend to ever have to do again: fiddle with these kind of settings in the future.
It reminds me of the days when Arnold just surfaced in the wild (no longer being a Sony-only in house renderer) where all it had was samples. Full stop. No thresholds, no tweaks or workarounds, no irradiance caches with dozens of settings that could be tweaked, no per-material samples tweaks. Just brute force samples. If itās noisy ā more samples, twice the samples twice the render time. But no more expensive artists spending 90% of their time setting render options like in V-Ray back then (and maybe even now? Havenāt used V-Ray since 10 years).
I have to leave for today, but hereās the rudimentary blend file I used. Youāll have to download the VDB for yourself though because itās several GBs large ( https://disneyanimation.com/resources/clouds/ ). But maybe you find settings that are visually identical or even better.
And remember, my laptop is old and slow. So if your compare the render times, compare them to Blender 5.0 on the same system.
Disney_Cloud_Blender45_v01.blend (110.9 KB)
