Page MenuHome

Mantaflow: Adaptive Domain breaks Dissolve if 'slow' is checked
Confirmed, NormalPublicBUG

Description

System Information
Operating system: Windows-10-10.0.18362-SP0 64 Bits
Graphics card: GeForce GTX 1070/PCIe/SSE2 NVIDIA Corporation 4.5.0 NVIDIA 441.28

Blender Version
Broken: version: 2.82 (sub 6), branch: master, commit date: 2020-01-18 18:58, hash: rB3e11c4e63b2f
Worked: (optional)

Short description of error
[Adaptive domain seems to completely ignore the 'dissolve' setting on the Domain object, or at least alter it massively.]

Exact steps for others to reproduce the error
[Bake the attached .blend
Here's the results-


The left hand Domain has dissolve enabled but not Adaptive. The right hand one is a duplicate of the left (and the left hand emitter), nothing at all changed apart from enabling Adaptive domain.
Doesn't look to me like the 'dissolve' parameter is doing terribly much at all on the right hand side, when 'Adaptive Domain' is enabled.]
[Based on the default startup or an attached .blend file (as simple as possible)]

I discovered this issue half way through a 400 frame render, when looking at the rendered stills; so I thought stuff-it! Carry on...
So here is what this issue does to an actual sim I wasted several hours on creating/tweaking, then the best part of a half a day rendering-

It looks worse than the .blend I uploaded as it has noise baked as well, and this seems to generally expand the volume filled by the smoke from what I've experienced, therefore exacerbating the jitter where the smoke hits the Adaptive Domain boundary.
As you can see it's quite ugly when the boundary of the smoke gets close to the camera, and as I hope you can see the jitter at the edges of the smoke look suspiciously like the jitter in the size of the domain seen in the viewport capture I made (obviously I matched the speed in Davinci, my workstation could only sustain 2FPS playback).
If anyone cares for my two-pennys-worth, whatever coded-wizardry controls the size of the Adaptive Domain, need's to implement some sort of 'damping' for the domain size changes, at least several (10-15?) frames worth. As well as taking 'dissolve' into account.

Event Timeline

Mark Spink (marks) added a comment.EditedJan 19 2020, 3:12 PM

And in case you were wondering...

It would appear to work as one would hope in the 'old' smoke sim...

Here's the blend file (2.81a not Mantaflow)-

Please, please, please don't kill the old smoke sim yet! It's still reliable...

Just tried it in today's build- version: 2.82 (sub 6), branch: master, commit date: 2020-01-19 22:44, hash: rB81b7f8efaf7a
Still the same.

Just tried it in version: 2.82 (sub 6), branch: master, commit date: 2020-01-20 22:06, hash: rB902209eda527
Still the same - but noticed a new issue-


Big ugly shadows(?) from the domain objects that don't show in the 'top orthographic' view, but do in Camera, User Perspective & Cycles render. I shall report separately.

Hello All, unfortunately this still happens in the Jan 23rd build-

Hello, still an issue in the Feb 3rd build.

Hello, still an issue in the Feb 4th build.

Germano Cavalcante (mano-wii) changed the subtype of this task from "Report" to "Bug".Thu, Feb 6, 2:01 PM

I cannot reproduce the problem.
But I'm seeing some uncomfortable glitches.
So I will confirm it one way or the other.

Operating system: Windows-10-10.0.18941 64 Bits
Graphics card: Radeon (TM) RX 480 Graphics ATI Technologies Inc. 4.5.13559 Core Profile Context 26.20.12028.2

@Mark Spink (marks), please simplify the description and make it clear what it is the "Short description of error" and what are the "Exact steps for others to reproduce the error".
Here are some tips for making a good bug report: https://wiki.blender.org/wiki/Process/Bug_Reports

Mark Spink (marks) added a comment.EditedThu, Feb 6, 3:02 PM

I don't get your question really, here's a screen capture of exactly what I've shown above in still images-

Recorded using the 4th feb build of 2.82Beta #2264590d8cc9.
Using the same scene as I've already uploaded, the left hand domain (no adaptive) shows the smoke dissolving/ending by 2 maybe 2.5 grid divisions below the X axis (after the initial plume which makes it all the way to the bottom of the domain). While in the domain on the right (with adaptive) the smoke makes it at least 4.5, if not 5 grid divisions below the X axis.

Clearly using adaptive domain either completely negates/ignores the 'dissolve' setting (same for both domains), or at best completely changes it's behaviour.
Maybe look from Top Ortho view, and look at any frame after say 50 (once the simulation has settled down), and try looking in 'look dev' (basically Eevee I'm guessing), like I show at the end of this capture.

Whatever your capture is showing, it's much worse/more than what I'm trying to report here. Maybe you have found another, different bug...

Mark Spink (marks) added a comment.EditedThu, Feb 6, 4:06 PM

Additionally: this still happens in the 5th Feb 2.82Beta build-

Interestingly, when re-baking the scene in the latest version (downloaded after my previous post), I tried un-checking the 'slow' option in dissolve, to see if this illustrated the issue even more obviously: not the case at all!

I would now like to amend the initial description from: Mantaflow: Adaptive Domain breaks Dissolve, to Mantaflow: Adaptive Domain breaks Dissolve if 'slow' is checked.
As you can see in this video, both domains behave pretty much the same if dissolve is used without the 'slow' checkbox enabled.

Looking at your previous screen capture again, I'm thinking this might be the difference between Cycles viewport playback using an RX480 as opposed a GTX 1070ti (with a second GTX1070- I'd guess Blender only uses the primary card for GUI/playback, but don't know for sure)
I had a quick google, and it looks like the 1070ti vs the RX480 is a bit like a 2008 VW Golf R32 vs a 2008 Golf 1.4: not quite twice as fast 0-60 (MPH), but almost.
https://gpu.userbenchmark.com/Compare/Nvidia-GTX-1070-Ti-vs-AMD-RX-480/3943vs3634
It could be the cache disk as well: I'm using 3 striped 1TB SSDs, and get about 1.4Gb/s read & write using Blackmagic Disk Speed test, could be a factor for viewport playback as well, assuming Blender doesn't cache the whole sim to memory (don't know I must admit).

Safest just to bake both domains in the scene, and look at a static frame after about frame 50, to see the issue (with 'slow' checked in 'dissolve'- as I now know)

Germano Cavalcante (mano-wii) changed the task status from Needs Triage to Confirmed.Fri, Feb 7, 4:42 PM

Just a heads-up from my side: @Mark Spink (marks) thanks for describing the issue so thoroughly! I will look into this again as soon as possible!

@Sebastián Barschkis (sebbas) Thanks! I shall be going dark until I install 2.82 release by the way (can't be arsed to download any more betas!)

Hello, this still happens in the 2.82 release build-

@Sebastián Barschkis (sebbas) here's a question: I'm assuming the 2.82 release build is effectively static, if this is the case- is there any way of knowing when the 2.83 alpha/beta build has been changed WRT Mantaflow?
Purely to know when it's worth downloading and trying for Mantaflow bugs, and not just randomly downloading the next newer build... (I am in Sydney and the internet out here is relatively bronze-age compared to most of the planet!)
(And I am English, so I'm supposed to moan about stuff in Aus as well)

Mark Spink (marks) renamed this task from Mantaflow: Adaptive Domain breaks Dissolve to Mantaflow: Adaptive Domain breaks Dissolve if 'slow' is checked.Sun, Feb 16, 1:03 AM

Hello, still happens in the 18th Feb build of 2.83alpha-