PDA

View Full Version : Relative particle age Broken in 2018



prometheus
01-21-2018, 11:46 AM
Hi folks!

I saw somewhere Kelly Myers was having issues with turbulenceFD, and decided to go hvīs for aeroplane smoke I think.
Thought I should have a go at such things again..Considering the old hypervoxels is much faster in 2018 than in 2015 (not talking the new volumetrics, that one are horribly slow with particles unfortunately)

set up a particle emitter parented to an aeroplanes engine, and moved it...added a particle emitter for the legacy hypervoxels, and was trying to use relative particle age for some channels in hypervoxels, but they do not seem to work unfortunately, the normal particle age works, but not relative particle age.

I also saw someone using it for liquid particles, and he is having issues of that not working either.

Bug?
And a report perhaps...

prometheus
01-21-2018, 12:31 PM
Probably be using it the wrong way, I think it may have must have been due to having set the particle emitter age to zero to keep particles running for whatever length the scene is, so once I changed the emitters particle age life to the same amount as the scene, the relative age now worked, at least with nodes and the new volumetric system, when I used blackbody radiator for getting that fire scatter look.

JamesCurtis
01-21-2018, 04:19 PM
Images and setup when you get a chance?

prometheus
01-21-2018, 05:07 PM
Images and setup when you get a chance?

As I mentioned, problem solved, it was particles not set right to zero life age, i donīt think neither hypervoxels or new volumetrics and any relative particle age could gather data in such case, see this thread, and the aeroplane image...
that is correct in the image...but there are other issues with the new volumetrics that is a bottleneck....osry for cross posting this pic..
This is the new volumetrics, I tried the black body radiator with old legacy hypervoxels,but it doesnīt work it seems, so here is the new volumetric system, unfortunately..adding any textures to the volumes if itīs applied on particles, it almost freezes the VPR.

and we miss random scale.

http://forums.newtek.com/showthread.php?155488-Hypervoxels-VS-New-Volumetrics/page7


http://forums.newtek.com/attachment.php?attachmentid=139647&d=1516568281

Anttij77
01-21-2018, 11:44 PM
"We need random scale values here"
You can get random scale for particles from the particle emitter itself. There are several options there.

"...speed becomes horrible..."
Adjust the "Step Size". If each of your particles is 150m in size and your step size is 2m, that roughly translates to 75 raymarch samples per particle. You can start with 50% of the Radius, and tweak the Step Size and Volumetric Samples (from the lights), to get to a nice quality/speed.

prometheus
01-22-2018, 12:42 AM
"We need random scale values here"
You can get random scale for particles from the particle emitter itself. There are several options there.

"...speed becomes horrible..."
Adjust the "Step Size". If each of your particles is 150m in size and your step size is 2m, that roughly translates to 75 raymarch samples per particle. You can start with 50% of the Radius, and tweak the Step Size and Volumetric Samples (from the lights), to get to a nice quality/speed.

Anti...I already mentioned this about random particle scale and change it from th particle emitter...I did that in my hypervoxel thread many days ago
..that is partly why I wrote this note in the image..it's a bad workflow compa red to old hypervoxels.

As for speed and step size...already tried that. ..but it's no good...The quality become s too bad..not acceptable.

prometheus
01-22-2018, 12:57 AM
To add...I must check
So there wasn't any background process interfering with the speed...
The old hypervoxels in 2018
Though it runs in circles around the new volumetrics..

Anttij77
01-22-2018, 05:50 PM
Though it runs in circles around the new volumetrics..

No, it really, really doesn't. Especially when you start making volumetrics that are not insanely dense (which is the default in the old), and when you try and fly through the volumetrics.
Also you simply cannot make realistic shading in the old, because it doesn't have realistic scattering and absorption values or the asymmetry, and the new one can react to GI and the old one simply can't.

Like this: https://upload.wikimedia.org/wikipedia/commons/a/af/Black_forest_fire_smoke_rising_to_sky.jpg

This is a render comparing apples to apples, as far as it can...
139669

Not only does the old look wrong, it was also a chore to even try to match the realistic look with the settings in the old, it also renders slower than the new.

Now let's try and fly thru the particle cloud, and see what happens... In this render I moved the camera to the center of the particle cloud. All the other settings are the same.
139670

prometheus
01-22-2018, 09:25 PM
No, it really, really doesn't. Especially when you start making volumetrics that are not insanely dense (which is the default in the old), and when you try and fly through the volumetrics.
Also you simply cannot make realistic shading in the old, because it doesn't have realistic scattering and absorption values or the asymmetry, and the new one can react to GI and the old one simply can't.

Like this: https://upload.wikimedia.org/wikipedia/commons/a/af/Black_forest_fire_smoke_rising_to_sky.jpg

This is a render comparing apples to apples, as far as it can...
139669

Not only does the old look wrong, it was also a chore to even try to match the realistic look with the settings in the old, it also renders slower than the new.

Now let's try and fly thru the particle cloud, and see what happens... In this render I moved the camera to the center of the particle cloud. All the other settings are the same.
139670

On my side. .hypervoxels is faster..
Firstly. .you did this test in 2018
For both of them right?
Secondly...
You need to use textures...doesn't seem you did that.
Third scale up the voxel to more than hundred meters.

Then this.
I do not compare full texture and shadow mode. .In often turn that off...so it may not be fair to compare them like that. ..but it can produce quite nice results. .All though not fully realistic gi.
Not always that is wanted. .apart from slowing down render time. .It's hard to avoid ..noise and it requires upping light volume samples.
But scale the volumetrics up to more real scale...and put textures on it.

Can't try now...I'm in bed.

01-23-2018, 08:54 AM
That's funny...
Sad, too....

prometheus
01-23-2018, 10:13 AM
That's funny...
Sad, too....

Dont make any gear overdrive conclusions on my findings..I Need to check it more carefully to verify some things. ..and also get back with a better description and summary of various different cases.

I would be happy to see more folks trying the new volumetrics out...but currently I seem to be the only one responding to it here on the forums..except for anti...from which I am greatful off that he checks it all out.

I am curious though why the slow down appears when textures are applied..and not when I only use spheres...particle s

jboudreau
01-23-2018, 02:06 PM
On my side. .hypervoxels is faster..
Firstly. .you did this test in 2018
For both of them right?


From the looks of it he used 2015.3 for one and 2018 for the other. Check out the icons on the renders.

Thanks,
Jason

jboudreau
01-23-2018, 02:57 PM
Dont make any gear overdrive conclusions on my findings..I Need to check it more carefully to verify some things. ..and also get back with a better description and summary of various different cases.

I would be happy to see more folks trying the new volumetrics out...but currently I seem to be the only one responding to it here on the forums..except for anti...from which I am greatful off that he checks it all out.

I am curious though why the slow down appears when textures are applied..and not when I only use spheres...particle s

Hi prometheus

Do you have a test scene that you can share where I can do some test for you.

Thanks,
Jason

prometheus
01-23-2018, 03:50 PM
Hi prometheus

Do you have a test scene that you can share where I can do some test for you.

Thanks,
Jason

Iīm off to bed now, and will probably not have time until the weekend.
But if I get inspired, I may cook up something.

prometheus
01-23-2018, 03:59 PM
From the looks of it he used 2015.3 for one and 2018 for the other. Check out the icons on the renders.

Thanks,
Jason

Thanks..I just noticed, and that is what I think I said specificly, do not compare hypervoxels in 2015 with the new volumetrics in 2018, I said...do compare hypervoxels legacy in 2018 with the new volumetrics in 2018, thatīs where hypervoxels is really fast..so Antis comparison isnīt correct when he talks about render speed, he compares two different lightwav versions..

and I also said you can use several nulls with hvs now, and design your clouds quite easy, in fact I think I sometimes can get better looking cloud to some degree with old hypervoxels legacy in 2018, than with the new system, at least it goes much faster to set things up..and render, then it of course lack the assymmetry, and gi illumination and the soft edges the new system has, but on the other hand, I think applying textures and controlling density is much easier...and compared to the new system..quite blistering fast.

when I look through content for clouds in 2018, it seems everything is on low meter scale, and once you try to scale it up, you of course need to raise step size or it will clog the renderer, but doing so yield quite nasty quality with almost real life scale clouds, now hypervoxels legacy do not have any such quality issue, it scales the same.

So a lot that worries me..
- destruction of a perfectly good workflow fo adjusting size variation (now only in particle emitters settings, or using nodes)
-destruction of a perfectly good workflow on how to access and add hypertextures, (now only with nodes) takes longer time to set up.
- large scale clouds doesnīt seem to hold up in quality and render speed.
-volumetrics can not be applied to any point clusters, only particles or nulls.
-no hypertexture motion effects, (need to add null refs)
- no good copy and paste function as we can do with hypervoxels.
- no turn off on function ..except entering scene editor.
- no presets ( I think..not sure)
- standing on the volumetric item in question is required for editing them, in hypervoxels you could move other items around and still having hypervoxels open to edit things, easier to jump back in again and edit a certain item than doing it from the objects tab and enter properties etc.

Hypervoxels specialized windows allowed for a gathered central control in a much motr user friendly way..that is what made it unique.
Oh..and we miss sprites, sprit clips, and what are we supposed to use for liquids surface modes for particles now...add hypervoxels?, if so they should have made hypervoxels easier to apply back again, till they implement a new liquid type of technique for particles o liquid fluids.